Machine Learning Approaches for Outdoor Air Quality Modelling: a Systematic Review

applied sciences

Review Machine Learning Approaches for Outdoor Air Quality Modelling: A Systematic Review

Yves Rybarczyk 1,2 and Rasa Zalakeviciute 1,* 1 Intelligent & Interactive Systems Lab (SI2 Lab), Universidad de Las Américas, 170125 Quito, Ecuador; [email protected] 2 Department of Electrical Engineering, CTS/UNINOVA, Nova University of Lisbon, 2829-516 Monte de Caparica, Portugal * Correspondence: [email protected]; Tel.: +351-593-23-981-000

Received: 15 November 2018; Accepted: 8 December 2018; Published: 11 December 2018

Abstract: Current studies show that traditional deterministic models tend to struggle to capture the non-linear relationship between the concentration of air pollutants and their sources of emission and dispersion. To tackle such a limitation, the most promising approach is to use statistical models based on machine learning techniques. Nevertheless, it is puzzling why a certain algorithm is chosen over another for a given task. This systematic review intends to clarify this question by providing the reader with a comprehensive description of the principles underlying these algorithms and how they are applied to enhance prediction accuracy. A rigorous search that conforms to the PRISMA guideline is performed and results in the selection of the 46 most relevant journal papers in the area. Through a factorial analysis method these studies are synthetized and linked to each other. The main ﬁndings of this literature review show that: (i) machine learning is mainly applied in Eurasian and North American continents and (ii) estimation problems tend to implement Ensemble Learning and Regressions, whereas forecasting make use of Neural Networks and Support Vector Machines. The next challenges of this approach are to improve the prediction of pollution peaks and contaminants recently put in the spotlights (e.g., nanoparticles).

Keywords: atmospheric pollution; predictive models; data mining; multiple correspondence analysis

1. Introduction Worsening air quality is one of the major global causes of premature mortality and is the main environmental risk claiming seven million deaths every year [1]. Nearly all urban areas do not comply with air quality guidelines of the World Health Organization (WHO) [2,3]. The risk populations that suffer from the negative effects of air pollution the most are children, elderly, and people with respiratory and cardiovascular problems. These health complications can be avoided or diminished through raising the awareness of air quality conditions in urban areas, which could allow citizens to limit their daily activities in the cases of elevated pollution episodes, by using models to forecast or estimate air quality in regions lacking monitoring data. Air pollution modelling is based on a comprehensive understanding of interactions between emissions, deposition, atmospheric concentrations and characteristics, meteorology, among others; and is an indispensable tool in regulatory, research, and forensic applications [4]. These models calculate and predict physical processes and the transport within the atmosphere [5]. Therefore, they are widely used in estimating and forecasting the levels of atmospheric pollution and assessing its impact on human and environmental health and economy [6–9]. In addition, air pollution modelling is used in science to help understand the relevant processes between emissions and concentrations, and understand the interaction of air pollutants with each other and with weather [10]

Appl. Sci. 2018, 8, 2570; doi:10.3390/app8122570 www.mdpi.com/journal/applsci Appl. Sci. 2018, 8, 2570 2 of 27 and terrain [11,12] conditions. Modelling is not only important in helping to detect the causes of air pollution but also the consequences of past and future mitigation scenarios and the determination of their effectiveness [4]. There are a few main approaches to air pollution modelling—atmospheric chemistry, dispersion (chemically inert species), and machine learning. Different complexity Gaussian models (e.g., AERMOD, PLUME) are widely used by authorities, industries, and environmental protection organizations for impact studies and health risk investigations for emissions dispersion from a single or multiple point sources (also line and area sources, in some applications) [13,14]. These models are based on assumptions of continuous emission, steady-state conditions and conservation of mass. Lagrangian models study a trajectory of an air parcel, the position and properties of which are calculated according to the mean wind data over time (e.g., NAME) [5,14]. On the other hand, Eulerian models use a gridded system that monitors atmospheric properties (e.g., concentration of chemical tracers, temperature and pressure) in specific points of interest over time (e.g., Unified Model). Chemical Transport Models (CTMs) (e.g., air-quality, air-pollution, emission-based, source-based, source, etc.) are prognostic models that process emission, transport, mixing, and chemical transformation of trace gases and aerosols simultaneously with meteorology [15]. Complex and computationally costly CTMs can be of a global (e.g., online: Fim-Chem, AM3, MPAS, CAM-Chem, GEM-MACH, etc.; and offline: GEOS-Chem, MOZART, TM5) and regional (e.g., online: MM5-Chem, WRF-Chem, BRAMS; and offline: CMAQ, CAMx, CHIMERE) scale [16,17]. These models combine atmospheric science and multi-processor computing techniques, highly relying on considerable resources like real-time meteorological data and an updated detailed emission sources inventory [18]. Unfortunately, emission inventory inputs for boundary layers and initial conditions may be lacking in some regions, while geophysical characteristics (terrain and land use) might further complicate the implementation of these models [19]. To deal with complex structure of air flows and turbulence in urban areas Computer Fluid Dynamics (CFD) methods are used [20]. However, recent studies show that the traditional deterministic models struggle to capture the non-linear relationship between the concentration of contaminants and their sources of emission and dispersion [21–24], especially in a model application in regions of complex terrain [25]. To tackle the limitations of traditional models, the most promising approach is to use statistical models based on machine learning (ML) algorithms. Statistical techniques do not consider physical and chemical processes and use historical data to predict air quality. Models are trained on existing measurements and are used to estimate or forecast concentrations of air pollutants according to predictive features (e.g., meteorology, land use, time, planetary boundary layer, elevation, human activity, pollutant covariates, etc.). The simplest statistical approaches include Regression [26], Time Series [27] and Autoregressive Integrated Moving Average (ARIMA) [28] models. These analyses describe the relationship between variables based on possibility and statistical average. Well-specified regressions can provide reasonable results. However, the reactions between air pollutants and influential factors are highly non-linear, leading to a very complex system of air pollutant formation mechanisms. Therefore, more advanced statistical learning (or machine learning) algorithms are usually necessary to account for a proper non-linear modelling of air contamination. For instance, Support Vector Machines [29], Artificial Neural Networks [30], and Ensemble Learning [31] have been applied to overcome non-linear limitations and uncertainties to achieve better prediction accuracy. Although statistical models do not explicitly simulate the environmental processes, they generally exhibit a higher predictive performance than CTMs on fin spatiotemporal scales in the presence of extensive monitoring data [32–34]. Different machine learning approaches have been used in recent years to predict a set of air pollutants using different combinations of predictor parameters. However, with a growing number of studies, it is puzzling why a certain algorithm is chosen over another for a given task. Therefore, in this study we aim to review recent machine learning studies used in atmospheric pollution research. To do so, the remainder of the paper is organized into three sections. First, we explain the method used to select Appl. Sci. 2018, 8, 2570 3 of 27 and scan relevant journal articles on the topic of machine learning and air quality, which conforms to the PRISMA guideline. This section also describes the strategy used to analyze the manuscripts and synthetizeAppl. Sci. the 2018 main, 8, x findings.FOR PEER REVIEW Second, the results are presented and discussed from a general to3 a of detailed 27 and synthetic account. Finally, the last section draws conclusions on the use of machine learning algorithms quality, which conforms to the PRISMA guideline. This section also describes the strategy used to for predictinganalyze the air qualitymanuscripts and theand future synthetize challenges the main of thisfindings. promising Second, approach. the results are presented and discussed from a general to a detailed and synthetic account. Finally, the last section draws 2. Method conclusions on the use of machine learning algorithms for predicting air quality and the future 2.1. Searchchallenges Strategy of this promising approach. Relevant2. Method papers were researched in SCOPUS. The enquiry was limited to this scientific search engine, because it compiles, in a single database, manuscripts which are also indexed in the 2.1. Search Strategy most significant databases in engineering (e.g., IEEE Xplore, ACM, etc.). The first step of the literature reviewRelevant consisted papers were of completingresearched in the SCOPUS.website The document enquiry search was withlimited a combinationto this scientific of search keywords. The usedengine, formula because was it compiles, as follows: in {‘Machinea single database Learning’}, manuscripts AND {‘Air which Quality’ are also OR indexed ‘Air Pollution’} in the most AND {‘Modelsignificant OR ‘Modelling’}. databases in The engineering exploration (e.g., was IEEE limited Xplore, to theACM, period etc.). ofThe 2010–2018. first step of Another the literature limitation review consisted of completing the website document search with a combination of keywords. The used was to focus only on journal papers, since they represent the most achieved work. The result of this formula was as follows: {‘Machine Learning’} AND {‘Air Quality’ OR ‘Air Pollution’} AND {‘Model first stepOR ‘Modelling’}. provides us The with exploration 103 documents. was limited to the period of 2010–2018. Another limitation was to Thefocus second only on step journal consisted papers, since of filtering they represent the studies, the most by achieved reading work. the title The and result the of abstract.this first step Papers were excludedprovides us from with our103 documents. selection if they addressed the topics as follows: Physical sensors (and not computationalThe second models); step health/epidemiological consisted of filtering the studiesstudies, (i.e.,by reading predictive the title models and the to estimateabstract. Papers the impact on healthwere and excluded not to from estimate our selection and/or if forecast they addressed the concentration the topics ofas pollutants);follows: Physical social sensors studies; (and biological not studies;computational indoor studies; models); sporadic health/epidemiological calamity (e.g., smog studies disaster). (i.e., predictive After applyingmodels to theseestimate rejection the impact criteria, the documentson health wereand not reduced to estimate to 50. and/or forecast the concentration of pollutants); social studies; Inbiological the last studies; step, all indoor 50 papers studies; were sporadic fully read.calamity After (e.g., a consensussmog disaster). between After theapplying authors these of this rejection criteria, the documents were reduced to 50. systematic review, four papers were rejected. Three manuscripts were excluded, because they In the last step, all 50 papers were fully read. After a consensus between the authors of this representedsystematic a very review, similar four study papers that were was previouslyrejected. Three carried manuscripts out by the were same excluded, authors. because The other they paper was consensuallyrepresented a consideredvery similar study out of that scope was afterpreviously reviewing carried the out full by the document. same authors. Consequently, The other paper a total of 46 manuscriptswas consensually were includedconsidered for out a furtherof scope qualitativeafter reviewing and the quantitative full document. synthesis. Consequently, Figure 1a totalrepresents of the flow46 manuscripts diagram ofwere the included search for method a further for qualitative the systematic and quantitative review. synthesis. It is based Figure on 1 represents the PRISMA approachthe flow that diagram provides of athe guideline search method to identify, for the select, system assessatic review. and summarizeIt is based on the the findingsPRISMA ofapproach similar but separatethat studies.provides Thisa guideline method to identify, promotes select, a quantitative assess and summarize synthesis the of findings the papers, of similar which but is separate carried out throughstudies. a factorial This method analysis. promotes a quantitative synthesis of the papers, which is carried out through a factorial analysis.

FigureFigure 1. PRISMA-based 1. PRISMA-based ﬂowchart flowchart of of the the systematicsystematic selection selection of of the the relevant relevant studies. studies. Appl. Sci. 2018, 8, 2570 4 of 27

2.2. Analyzed Parameters All the papers are analyzed according to 14 aspects. The first parameter concerns the motivation of the study. The second is the type of modelling, which is divided into two categories—estimation and forecasting models. An estimation model uses predictive features (e.g., contaminant covariates, meteorology, etc.) to estimate the concentration of a determined pollutant at the same time. A forecasting model takes into account historical data to predict the concentration of a pollutant in the future. The third analysis is based on the type of machine learning algorithm used. The main categories are artificial neural networks, support vector machines, ensemble learning, regressions, and hybrid versions of these algorithms. The fourth analysis describes the method applied by the authors. The fifth point focuses on the nature of the predicted parameter. Again, two groups are identified. On the one hand, there are authors interested in predicting specific air contaminants, which are: Micro and nano-size particulate matter (PM10, PM2.5, PM1); nitrogen oxides (NOx = NO + NO2); Sulphur oxides (SOx); carbon monoxide (CO); and ozone (O3). On the other hand, some authors work on a prediction of air quality in general by searching for the Air Quality Index (AQI), which may include the concentrations of several pollutants. The sixth parameter identifies the geographic location of the study. The seventh gives details of the characteristics of the dataset, such as—time span; quantity of monitoring stations; and number of instances. Furthermore, the eighth point provides information on the specificity of the dataset in terms of the used predictive attributes. The main features are related to—pollutant covariates; meteorology; land use; time; human activity; and atmospheric phenomena. The ninth and tenth factors address the evaluation method and the performance of the tested algorithms, respectively. The assessment is mainly based on a comparison of the accuracy between the models and/or a comparison of the prediction of the actual value. The most popular evaluation criteria are the ratio of correctly classified instances (Accuracy), the Mean Absolute Error (MAE), the Root Mean Square Error (RMSE), and the coefficient of determination (R2). The Accuracy represents the overall performance of a classifier by providing the proportion of the whole test set that is correctly classified, as described in Equation (1).

TP + TN Accuracy = (1) TP + TN + FP + FN where TP, TN, FP and FN stand for True Positives, True Negatives, False Positives and False Negatives, respectively. The higher the Accuracy value is, the better is the model performance. The MAE shows the degree of difference between the predicted values and the actual values. The RMSE is another relative error estimator that focuses on the impact of extreme values based on the MAE. The R2 represents the ﬁtting degree of a regression. The MAE, RMSE and R2 are calculated according to Equations (2)–(4), respectively.

1 n MAE = ∑|Ei − Ai| (2) n i=1

1 n ! 2 1 2 RMSE = ∑(Ei − Ai) (3) n i=1  2 n 2 ∑i=1 Ai − A Ei − E R =  q  (4) n 2 n 2 ∑i=1 Ai − A ∑i=1 Ei − E where n is the number of instances, Ai and Ei are the actual and estimated values, respectively. A and E stand for the mean measured and mean estimated value of the contaminant, respectively. Amax and 2 Amin are the maximum and minimum observed pollutant values, respectively. R is a dimensionless descriptive statistical index ranging from 0 (no correlation) to 1 (perfect correlation). MAE and RMSE Appl. Sci. 2018, 8, 2570 5 of 27 values are in # cm−3. The lower the MAE and RMSE is, the better is the predictive performance of the model. The eleventh parameter considers the computational cost of the method. Finally, the last three points discuss the outcomes of the proposed approach, its limitation, and its scope of applicability.

2.3. Synthesis of Results The results are synthetized according to descriptive statistics and a factorial analysis. First, we describe the most used algorithms over time in order to define the current tendencies. Second, we quantify the types of algorithms applied for the prediction of the principal contaminants. Third, we identify the evolution of the modelling performance for each pollutant over the last decade. Finally, since several parameters are considered for the description of the selected papers, we perform a factorial analysis to summarize the main outcomes of this systematic review. The most appropriate method to identify the relationships between the qualitative factors that characterize each study is the Multiple Correspondence Analysis (MCA). First, a data table is created from the parameters identified in Section 2.2 (see Figure A1 of AppendixA). Each row (i) corresponds to each paper and each column (j) corresponds to each qualitative variable (e.g., types of modelling, algorithms, etc.). The total number of papers and qualitative variables are identified as I and J, respectively. Next, this data table is fed to the R software, in order to proceed with the analysis through the MCA library. The initial table is transformed into an indicator matrix called the Complete Disjunctive Table, in which rows are still the papers, but the columns are now the categories (e.g., estimation model vs. forecasting model) of the qualitative variables. The total number of categories is identified as kj. The entry of the intersection of the i-th row and the k-th column, which is called yik, is equal to 1 (or true) if the i-th paper has category k of the j-th variable and 0 (or false) otherwise. All the papers have the same weight, which is equal to 1/I. However, the method highlights papers that are characterized by rare categories by implementing Equation (5).

yik xik = (5) pk where pk represents the proportion of papers in the category k. Then, the data are centralized by applying Equation (6).

yik xik = − 1 (6) pk

The table of the xik is used to build the points cloud of papers and categories. Since, the variables are centered, the cloud has a center of gravity in the origin of the axes. The distance between a paper i and a paper i0 is given by Equation (7).

K K 2 pk 2 1 1 2 di,i0 = ∑ (xik − xi0k) = ∑ (yik − yi0k) (7) k=1 J J k=1 pk 0 where xik and xi0k are the coordinates of the papers i and i , respectively. This equation shows that the distance is 0 if two papers are in the same set of categories (or proﬁle). If two papers share many categories, the distance will be small. However, if two papers share several categories except a rare one, the distance between them will be relatively large due to the pk value. Next, the calculation of the distance of a point from the origin (O) is given by Equation (8).

K K y 2 pk 2 1 ij d(i, O) = ∑ (xik) = ∑ − 1 (8) k=1 J J k=1 pk The equation shows that a distance gets larger when the categories of a paper are rarer, because its pk are small. In other words, the more a paper possesses rare categories, the further it is from the origin of the plot axes. To conclude the process, the point cloud must be projected into a smaller-dimensional Appl. Sci. 2018, 8, 2570 6 of 27 space, which is usually reduced to two dimensions. To do so, the cloud (N ) is projected onto a Appl. Sci. 2018, 8, x FOR PEER REVIEW I 6 of 27 sequence of orthogonal axes with maximum inertia as calculated in Equation (9).

I 11 2 InertiaInertia((NNI) ==maxmax dd ((i,i, O O) (9)(9) ∑ II i=1 In terms of the first first two dimensions obtained, we end up with a factor plane for the papers and categories that is the best bi-dimensionalbi-dimensional representation of them. 3. Results and Discussion 3. Results and Discussion This section is organized in three parts. First, a general description of the results of the systematic This section is organized in three parts. First, a general description of the results of the systematic review is provided. These results are mainly based on descriptive statistics regarding the number review is provided. These results are mainly based on descriptive statistics regarding the number of of studies that uses a machine learning approach over the last decade, the geographic distribution studies that uses a machine learning approach over the last decade, the geographic distribution of of these studies across the world, and most popular algorithms implemented. The second part these studies across the world, and most popular algorithms implemented. The second part provides provides a detailed development of the 46 papers selected following the method defined in Section 2.1. a detailed development of the 46 papers selected following the method defined in Section 2.1. These These papers are grouped into six categories for a more comprehensive description. The definition papers are grouped into six categories for a more comprehensive description. The definition of the of the groups is based on the principal motivation of the study. Category 1 groups the manuscripts groups is based on the principal motivation of the study. Category 1 groups the manuscripts that are that are interested in identifying the most relevant predictors and understanding their non-linear interested in identifying the most relevant predictors and understanding their non-linear relationship relationship with air pollution. Category 2 corresponds to a set of papers that deals with image-based with air pollution. Category 2 corresponds to a set of papers that deals with image-based monitoring monitoring and the way to tackle low spatial resolution from non-specific/low resolution sensors. and the way to tackle low spatial resolution from non-specific/low resolution sensors. Category 3 is Category 3 is specialized on a family of studies that considers land use and/or spatial heterogeneity specialized on a family of studies that considers land use and/or spatial heterogeneity as a feature. as a feature. Category 4 focuses on hybrid models as well as extreme and deep learning. Category 5 Category 4 focuses on hybrid models as well as extreme and deep learning. Category 5 is a bunch of is a bunch of work, which resulted in an applicative system. Finally, Category 6 addresses the new work, which resulted in an applicative system. Finally, Category 6 addresses the new challenge of challenge of predicting the concentration of nanoparticle matter. predicting the concentration of nanoparticle matter. 3.1. General Description 3.1. General Description Figure2 shows the evolution of the number of journal papers using a machine learning approach Figure 2 shows the evolution of the number of journal papers using a machine learning approach to model air quality, since 2010. The number of studies in this field shows an increasing trend over the to model air quality, since 2010. The number of studies in this field shows an increasing trend over last eight years. An important boom can be observed from 2015. The average number of studies has the last eight years. An important boom can be observed from 2015. The average number of studies been multiplied by a factor of 5, since the period 2011–2015. This high value tends to remain stable has been multiplied by a factor of 5, since the period 2011–2015. This high value tends to remain stable during the last two years. It is of note that the literature review was performed in September 2018, during the last two years. It is of note that the literature review was performed in September 2018, which supposes a possibly superior number of documents by the end of the present year. which supposes a possibly superior number of documents by the end of the present year.

Figure 2. TheThe evolution evolution of of published published studies studies in in scientific scientiﬁc journals regarding machine learning techniques used in atmospheric sciences.

While the number of machine learning approaches for predicting atmospheric pollutants has increased dramatically, this growth has not been uniformed globally. Figure3 shows that the number of studies is much higher in the northern hemisphere, specifically in Eurasia and North America. In addition, the more complex list of pollutants has been studied in Asia and Europe, including Appl. Sci. 2018, 8, x FOR PEER REVIEW 7 of 27 all criteria pollutants and even a specific pollutant or an overall Air Quality Index (AQI). While the studies),biggest number other parts of studies of the has world been have concentrated more limite ind China published (13 studies) research. and Especially United States unstudied (seven studies),regions ofother the partsworld of are the South world America have more and limited Oceania, published both with research. one study, Especially focusing unstudied on a single regions specific of the worldpollutant. are South America and Oceania, both with one study, focusing on a single specific pollutant.

Figure 3. NumberNumber of of papers papers per per country country and and types types of of atmospheric atmospheric contaminants contaminants studied studied by by continent. continent. The world map is colored by number of studies per country (yellow–brown), while the grey color

indicates regions with no studies.studies. The pollutant studies werewere consideredconsidered perper continent:continent: OO3—ozone, PM10—particulate—particulate matter with aerodynamic diametersdiameters ≤≤1010 µm,µm, BC—black carbon, PM2.52.5—particulate—particulate matter with with aerodynamic aerodynamic diameters diameters ≤2.5≤ 2.5µm,µ AQI—Airm, AQI—Air Quality Quality Index, Index, NO2—nitrogen NO2—nitrogen dioxide, dioxide, NOx— nitrogenNOx—nitrogen oxides oxides(NO + (NONO2), + SO NO2—sulfur2), SO2 —sulfurdioxide, dioxide,CO—carbon CO—carbon dioxide, dioxide,NanoPM—nano NanoPM—nano particles particles(particles(particles with diameter with diameterin nm). in nm).

The systematic review shows that the most used algorithms in descending order are: Ensemble Learning Methods, Artificial Artificial Neural Networks, Su Supportpport Vector Machines, and Linear Regressions (Figure 44).). TwoTwo paperspapers useuse moremore peculiarpeculiar approachesapproaches relatedrelated toto DecisionDecision rulesrules andand LazyLazy methods.methods. Ensemble Learning Learning is is a a class class of of machine machine learning learning techniques techniques that that creates creates multiple multiple predictors predictors to addressto address the same the sameproblem problem and make and a single make prediction a single predictionby combining by the combining results from the the results predictors. from Thus,the predictors. the result of Thus, the predictive the result model of the is predictive obtained modelby taking is obtainedan average by or taking the majority an average voting. or The the multiplemajority predictors voting. The are multiple created predictors by introducing are created stocha bysticity introducing into the data stochasticity or the prediction into the dataalgorithm. or the predictionIn the first case, algorithm. different In thedatasets first case, and differentrespective datasets predictors and are respective produced predictors from the are original produced database. from Inthe the original second database. case, the randomness In the second is case,introduced the randomness by using different is introduced types of by algorithms using different (e.g., Decision types of Treesalgorithms and Neural (e.g., DecisionNetwork) Trees to solve and a Neural single proble Network)m. The to solveadvantage a single of using problem. an Ensemble The advantage Learning of methodusing an is Ensemble to provide Learning a better method accuracy, is to if provide we compare a better to accuracy, the performance if we compare of each to thepredictor performance taken individuallyof each predictor (also taken called individually weak predictors). (also called Bagging weak (or predictors). Bootstrap Aggregation) Bagging (or Bootstrap is the simplest Aggregation) way to introduceis the simplest stochasticity way to in introduce the data. stochasticity In this techniqu in thee, several data. Indatasets this technique, are created several by taking datasets random are samplescreated bywith taking replacement. random samples Then, one with specific replacement. predictor Then, is produced one specific for predictoreach dataset, is produced and theses for predictorseach dataset, are and aggregated theses predictors to give the are final aggregated model. B tooosting give theis another final model. way to Boosting create stochasticity is another way by sequentiallyto create stochasticity training each by sequentially weak classifier. training These each predictors weak classifier. are weighted These in predictors some way are that weighted is usually in related to the accuracy of each predictor. It means that observations which are incorrectly predicted get a higher weight. So, the best predictor is the one that has to perform better on these critical observations. Finally, Stacking is the method used to make a prediction from a battery of different possible algorithms. After applying these algorithms to the dataset, they are assessed by cross- validation, for instance, in order to give a weight to each method according to its performance. Random Forest is one of the most popular Ensemble Learning techniques and constitutes the majority of the papers classified under this label in our literature review. This method is based on an ensemble Appl. Sci. 2018, 8, 2570 8 of 27 some way that is usually related to the accuracy of each predictor. It means that observations which are incorrectly predicted get a higher weight. So, the best predictor is the one that has to perform better on these critical observations. Finally, Stacking is the method used to make a prediction from a battery of different possible algorithms. After applying these algorithms to the dataset, they are assessed by cross-validation, for instance, in order to give a weight to each method according to its Appl.performance. Sci. 2018, 8, x Random FOR PEER Forest REVIEW is one of the most popular Ensemble Learning techniques and constitutes8 of 27 the majority of the papers classified under this label in our literature review. This method is based on ofan Decision ensemble Tree of Decision predictors. Tree The predictors. first step The implem first stepents implementsbagging on baggingtrees. Then, on trees. a second Then, step a second adds stochasticitystep adds stochasticity to split the totree split by theusing tree the by random using the subspace random method subspace or methodthe random or the split random selection, split whichselection, consists which of consistsapplying of at applying each node at eachthe algorithm node the algorithmwith a subset with only a subset of the only features of the to features split the to node.split the node.

FigureFigure 4. 4. ProportionProportion of machine of machine learning learning algorithms algorithms used. ‘Ensemble’ used. ‘Ensemble’ stands for stands Ensemble for EnsembleLearning (mainlyLearning Random (mainly Forest). Random ‘NN’ Forest). stands for ‘NN’ any standsalgorithm for based any algorithmon an Artificial based Neural on an Networks Artificial approach.Neural Networks ‘SVM’ stands approach. for Support ‘SVM’ Vector stands Machine for Support (it also Vector includes Machine Support (it also Vector includes Regression). Support Vector Regression). ‘Regression’ stands for Multiple Linear Regression. ‘Others’ includes Decision ‘Regression’ stands for Multiple Linear Regression. ‘Others’ includes Decision Rules and Lazy Rules and Lazy methods. methods. Artificial Neural Networks (NN) are inspired by the working of the human being’s nervous system. Artificial Neural Networks (NN) are inspired by the working of the human being’s nervous The simplest neural network architecture is called Perceptron. This method classifies inputs by using a system. The simplest neural network architecture is called Perceptron. This method classifies inputs linear combination of the observations of the dataset, as defined in Equation (10). by using a linear combination of the observations of the dataset, as defined in Equation (10). J x = w o (10) x= w∑ j oj (10) j=0 where o is an observation characterized by several features (from 0 to J) and wj is the weight applied to where o is an observation characterized by several features (from 0 to J) and wj is the weight applied each feature. For instance, in the case of two classes, if x is greater than 0 the observation is classified to each feature. For instance, in the case of two classes, if x is greater than 0 the observation is classified in class 1, otherwise it is classified in class 2. In an NN, the weights are learned. The algorithm starts in class 1, otherwise it is classified in class 2. In an NN, the weights are learned. The algorithm starts by setting all weights to 0. Then, a loop is applied until all observations in the training data are by setting all weights to 0. Then, a loop is applied until all observations in the training data are classified correctly. In this loop, for each observation o, if o is classified incorrectly and if it belongs to classified correctly. In this loop, for each observation o, if o is classified incorrectly and if it belongs class 1, it is added to the weight vector, otherwise it is subtracted. No modification is applied if the to class 1, it is added to the weight vector, otherwise it is subtracted. No modification is applied if the observation is classified correctly. Nevertheless, the NN implemented in the selected papers consist of observation is classified correctly. Nevertheless, the NN implemented in the selected papers consist at least three types of layers (i.e., Multilayer Perceptron)—input layer, hidden layer, and output layer. of at least three types of layers (i.e., Multilayer Perceptron)—input layer, hidden layer, and output The input is composed with the features (e.g., contaminant covariates, meteorology, etc.). The output layer. The input is composed with the features (e.g., contaminant covariates, meteorology, etc.). The is the variable to be predicted (i.e., concentration of a contaminant or value of an air quality index). output is the variable to be predicted (i.e., concentration of a contaminant or value of an air quality And the hidden layer consists of additional nodes between the two previous layers that enables index). And the hidden layer consists of additional nodes between the two previous layers that multiple kinds of connections. When the number of hidden layers is large the NN is called Deep enables multiple kinds of connections. When the number of hidden layers is large the NN is called Learning Neural Network. Each node (or neuron) performs a weighted sum of its input and thresholds Deep Learning Neural Network. Each node (or neuron) performs a weighted sum of its input and thresholds the result. Instead of using the binary threshold of the basic Perceptron, the Multilayer Perceptron usually applies a smooth activation function based on a sigmoid. In such an implementation, as the inputs become more extreme, they approach the step function, which corresponds to the hard-edge threshold used in the basic Perceptron. Finally, the learning stage consists of getting the best weights to minimize the error between the predictive output and the actual output. This is an iterative process using a steepest descent method, in which the gradient is determined by a backpropagation algorithm that seeks to minimize the cost function J(θ) defined in Equation (11). Appl. Sci. 2018, 8, 2570 9 of 27 the result. Instead of using the binary threshold of the basic Perceptron, the Multilayer Perceptron usually applies a smooth activation function based on a sigmoid. In such an implementation, as the inputs become more extreme, they approach the step function, which corresponds to the hard-edge threshold used in the basic Perceptron. Finally, the learning stage consists of getting the best weights to minimize the error between the predictive output and the actual output. This is an iterative process using a steepest descent method, in which the gradient is determined by a backpropagation algorithm that seeks to minimize the cost function J(θ) defined in Equation (11).

m 2 1 (i) (i) J(θ) = ∑ hθ x − y (11) m i=1 where m is the number of observations, y is the actual output and hθ(x) is the predicted output. Support Vector Machine (SVM) is another algorithm that draws a boundary through the widest channel between two classes, which is the maximum separation from each class. This boundary line is obtained by selecting the critical points (or support vectors) that deﬁne the channel and taking the perpendicular bisector of the line joining those two support vectors. In the case of a multidimensional dataset the boundary is deﬁned by a hyperplane. Equation (12) shows that the calculation of the maximum margin hyperplane depends only on the value of the support vectors and not the remaining observations.

I α x = b + ∑ iyia(i) × a (12) i=0 where a(i) are the support vectors and I is the number of these vectors. In cases where classes are not linearly separable, a function is used to transform the data into a high dimensional space (e.g., a polynomial function). A kernel trick is used to reduce the computational cost of this transformation. The most current kernel types are —linear kernel (Equation (13)), polynomial kernel (Equation (14)), and radial basis function kernel (Equation (15)).

K(x, y) = x × y (13)

K(x, y) = (x × y + 1)d (14)

2 K(x, y) = e−γ||x−y|| (15)

It is important to choose the right kernel and to tune its parameters correctly to get a good performance from a classifier. A usual parameter tuning technique includes k-fold cross-validation. SVM is a popular alternative to artificial neural networks. Finally, a fair number of papers applies a Regression method as machine learning algorithm to predict air quality. Features derived from the dataset are used as input of the Regression model to predict continuous valued output. As in the NN approach, the prediction is obtained by learning the relationship (or weights) between the inputs and the output. These weights are acquired by fitting a linear or nonlinear curve to the data points. In order to correctly fit the curve, it is necessary to define the goodness-of-fit metric, which allows us to identify the curve that fits better than the other ones. The optimization technique used in regression, and in several other machine learning methods (e.g., NN), is the gradient descent algorithm that aims to minimize the cost function J(θ) defined in Equation (11).

3.2. Detailed Description This section consists of a detailed explanation of the selected papers. It is organized into six categories based on the main motivation of the study. In each group, we ﬁrst present manuscripts on estimation models and, second, studies that involve forecasting models. An estimation model is deﬁned as an approach that seeks to estimate the value of a pollutant (or index of contamination) from Appl. Sci. 2018, 8, 2570 10 of 27 the measurement of other predictive parameters at the same time step. It is mostly used to give an approximation of the concentrations on an extended geographic area. On the contrary, the objective of a forecasting model is to predict the concentration levels of the contaminants in the near future. We call both estimation and forecasting prediction or predictive models.

3.2.1. Category 1: Identifying Relevant Predictors and Understanding the Non-Linear Relationship with Air Pollution Category 1 is the biggest group and accounts for total of 16 articles—twelve in estimation modeling and four in forecasting. In the estimation modeling there are five cases of Random Forest (RF) or tree-based Ensemble learning algorithms (e.g., M5P), four regression models, one lazy learning, and two SVM. The latter was once used in forecasting studies, NN was used twice and M5P once. This category highly concentrates on identifying the most contributing parameters to a successful prediction of atmospheric pollution. A variety of tests are used to differentiate between the influential predictors, like Principal Component Analysis (PCA), Quantile Regression Model (QRM), Pearson correlation, etc. Among the most recent studies we find Grange et al. [22]. This study aims to (i) build a predictive model of PM10 based on meteorological, atmospheric, and temporal factors and (ii) analyze PM10 trend during the last ten years. The Random Forest model, a simple, efficient, and easy interpretable method, is used. First of all, many out-of-bag samples (randomly sampling observations with replacement and sampling of feature variables) of the training set are used to grow different Decision Trees (DT), and then all the trees are aggregated to form a single prediction. The algorithms are trained and run on 20 years of data in Switzerland from 31 monitoring stations. Daily average data of meteorology (wind speed, wind direction, and temperature), synoptic scale, boundary layer height, and time are used. The model is validated by comparison between the observed and predicted value, resulting in average value in 31 sites: r2 = 0.62. The best predictors are—wind speed, day of the year (seasonal effect), and boundary layer height, whereas the worst predictors are day of the week, and synoptic scale. The model performance is low and the predictive accuracy is inconsistent (0.53 < r2 < 0.71) for different locations, especially varying for the rural mountain sites. Another study also relies on RF to build a spatiotemporal model to predict O3 concentration across China [35]. First the RF model is built by averaging the predictions from 500 decision/regression trees, and then Random Forest Regressor from the python package scikit-learn is used to run the algorithm. For that meteorology, planetary boundary height, elevation, anthropogenic emission inventory, land use, vegetation index, road density, population density, and time are used from 1601 stations over one year in all of China. The model is validated by comparison of the r2 and RMSE (predicted value/actual value) to Chemical Transport Models, and shows a good performance of r2 = 0.69 (specifically, r2 = 0.56 in winter, and r2 = 0.72 in fall). Interestingly, the model results are quite comparable or even higher than the predictive performance of the CTMs for lower computational costs. While for the predictive features, meteorological factors account for 65% of the predictive accuracy (especially humidity, temperature, and solar radiation), also showing a higher predictive performance when the weather conditions are stronger (i.e., autumn). In this study, anthropogenic emissions (NH3, CO, Organic Carbon, and NOx) exhibit a lower importance than meteorology and lower accuracy is registered for the regions with a sparser density of monitoring stations. Therefore, the accuracy relies on the complexity of the network. Martínez-España et al. [36] aim to identify the most robust machine learning algorithms to preserve a fair accuracy when O3 monitoring failures may occur, by using five different algorithms—Bagging vs. Random Committee (RC) vs. Random Forest vs. Decision Tree vs. k-Nearest Neighbors (kNN). First, the prediction accuracy of the five machine learning algorithms is compared, then a hierarchical clustering technique is applied to identify how many models are needed to predict the O3 in the region of Murcia, Spain. Two years of pollutant covariates (O3, NO, NO2, SO2, NOx, PM10,C6H6,C7H8, and XIL) and meteorological parameters (temperature, relative humidity, pressure, Appl. Sci. 2018, 8, 2570 11 of 27 solar radiation, wind speed and direction) are used. The model is validated by the comparison of the coefficient of determination (predicted value/actual value) between the five models. Random Forest slightly outperforms the other models (r2 = 0.85), followed by RC (r2 = 0.83), Bagging and DT (r2 = 0.82) 2 and kNN (r = 0.78). The best predictors are NOx, temperature, wind direction, wind speed, relative humidity, SO2, NO, and PM10. Finally, the hierarchical cluster shows that two models are enough to describe the studied regions. Bougoudis et al. [37] aim to identify the conditions under which high pollution emerges and to use a better generalized model. Hybrid system based on the combination of unsupervised clustering, Artificial Neural Networks (ANN) and Random Forest (RF) ensembles and fuzzy logic is used to predict multiple criteria pollutants in Athens, Greece. Twelve years of hourly data of CO, NO, NO2, SO2, temperature, relative humidity, pressure, solar radiation, wind speed and direction are used. Unsupervised clustering of the initial dataset is executed in order to re-sample the data vectors; while ensemble ANN modeling is performed using a combination of machine learning algorithms. The optimization of the modeling performance is done with Mamdani rule-based fuzzy inference system (FIS, used to evaluate the performance of each model) that exploits relations between the parameters affecting air quality. Specifically, self-organizing maps are used to perform dataset re-sampling, then ensembles of feedforward artificial neural networks and random forests are trained to clustered data vectors. The estimation model performance is quite good for CO (r2 = 0.95, RF_FIS), 2 2 2 NO (r = 0.95, ensemble NN_FIS), NO and O3 (r = 0.91, RF) and SO2 (r = 0.78, ensemble regression). Sayegh et al. [38] also employ a number of models (Multiple Linear Regression vs. Quantile Regression Model (QRM) vs. Generalized Additive Model vs. Boosted Regression Trees) to perform a comparative study on the performance for capturing the variability of PM10. At the contrary of the linear regression, which considers variables distribution as a whole, the QRM defines the contribution (coefficient) of the predictors for different percentiles of PM10 (here 10 percentiles or quantiles are used) to estimate the weight of each feature. Meteorological factors (wind speed, wind direction, temperature, humidity), chemical species (CO, NOx, SO2), and PM10 of the previous day data for one year from Makkah (Saudi Arabia) are used. The model performance is validated with 2 observed data and the QRM model shows a better performance (r = 0.66) in predicting hourly PM10 concentrations due to the contribution of covariates at different quantiles of the dependent variable (PM10), instead of considering the central tendency of PM10. Singh et al. [39] aim to identity sources of pollution and predict the air quality by using Hybrid Model of Principal Components Analysis, Tree-based Ensemble Learning (Single Decision Tree (SDT), Decision Tree Forest (DTF) and Decision Tree Boost (DTB)) vs. Support Vector Machine Model. Five years of air quality and meteorological parameter data are used for Lucknow (India). SDT, DTF, DTB and SVM are used to predict the Air Quality Index (AQI), and Combined AQI (CAQI)), and to determine the importance of predictor features. The model performance is validated by a comparison between the models and with the observed data. Decision Tree models—SDT (r2 = 0.9), DTF (r2 = 0.95), and DTB at r2 = 0.96) outperform SVM (r2 = 0.89). Philibert et al. [40] aim to predict a greenhouse gas N2O emission using Random Forest vs. Two Regression models (linear and nonlinear) by employing global data of environmental and crop variables (e.g., fertilization, type of crop, experiment duration, country, etc.). The extreme values from data are removed, excluding the boreal ecosystem; input variable features are ranked by importance; controlling number of input variables to result in better prediction. The model is validated by a comparison to the regression model and the simple non-linear model (10-fold cross validation). RF outperforms the regression models by 20–24% of misclassifications. Meanwhile, Nieto et al. [41] aim to predict air pollution (NO2, SO2 and PM10) in Oviedo, Spain based on a number of factors using Multivariate Adaptive Regression Splines and ANN Multilayer Perceptron (MLP). Three years of NOx, CO, SO2,O3, and PM10 data modeling result in a good 2 2 2 estimation of NO2: r = 0.85; SO2: r = 0.82; PM10 r = 0.75 when compared with observed data. Appl. Sci. 2018, 8, 2570 12 of 27

Kleine Deters et al. [42] aim to identify the meteorology effects on PM2.5, by predicting it using six years of meteorological data (wind speed and direction, and precipitation) for Quito, Ecuador in regression modeling. This is a good simplified technique and an economic option for the cities with no air quality equipment. The model gets validated by regression between observed and predicted PM2.5, and by a 10-fold cross validation of predicted vs. observed PM2.5, which shows that this method is a fair approach to estimate PM2.5 from meteorological data. In addition, the model performance is improved in the more extreme weather conditions (most influential parameters), as previously seen in Zhan et al. [35]. Carnevale et al. [43] use the lazy learning technique to establish the relationship between precursor emissions (PM10) and pollutants (AQI of PM10) for the Lombard region (Italy), using hourly Sulfur Dioxide (SO2), Nitrogen oxides (NOx), Carbon Monoxide (CO), PM10, NH3 data for one year. The Dijkstra algorithm is deployed in the large-scale data processing system. Eighty percent of the data are used as an example set and 20% as validation set (every 5th cell). The validation phase of surrogate models is performed comparing the output to deterministic model simulations, not the observations. Constant, linear, quadratic and combination approximation based on elevation of the area (0, 1, 2 and all polynomial approximations) are applied. The performance of the model is very comparable (r2 = 0.99–1) to the Transport Chemical Aerosol Model (TCAM), a model that is computationally much costlier, currently used in decision making. Suárez Sánchez et al. [44] on the other hand, aim to estimate the dependence between primary and secondary pollutants; and most contributing factors in air quality using SVM radial (Gaussian), linear, quadratic, Pearson VII Universal Kernels (PUK) and multilayer perceptron (MLP) to predict NOx, CO, SO2,O3, and PM10. Three years of NO, NO2, CO, SO2,O3, and PM10 data are used in Aviles, Spain. The model is 10-fold cross validated with the observed data, resulting in best performance using PUK 2 for NOx and O3 (r = 0.8). Liu et al. [23] also employ SVM to get the most reliable predictive model of air quality (AQI) by considering monitoring data of three cities in China (Beijing, Tianjin, and Shijiazhuang). Two years of last-day AQI values, pollutant concentrations (PM2.5, PM10, SO2, CO, NO2, and O3), meteorological factors (temperature, wind direction and velocity), and weather description (ex. cloudy/sunny, or rainy/snowy, etc.) are used as predicting features. The dataset is split into a training set and testing set through a 4-fold cross validation technique. To validate the model performance, the results are compared to the observed data. The model performance is especially improved when the surrounding cities’ air quality information is included. Vong et al. [45] also use SVM to forecast air quality (NO2, SO2,O3, SPM) from pollutants and meteorological data in Macau (China). The training is performed on three years of data and tested on one whole year and then just on January and July for seasonal predictions. The Pearson correlation is used to identify the best predictors for each pollutant and different kernels are used to test which of the predictors or models get the best results. The Pearson correlation is also employed to determine how many days of data are optimal for forecasting. The model performance is validated by comparison to the observed data resulting in best kernel—linear and RBF, in polynomial (summer), confirming that SVM performance depends on the appropriate choice of the kernel. In the forecasting model category, there are two articles that use NN, and meteorology and pollutants as predicting features. Chen et al. [46] aim to build a model that forecasts AQI one day ahead by using an Ensemble Neural Network that processes selected factors using PEK-based machine learning for 16 main cities in China (three years of data). First a selection of the best predictors (PM2.5, PM10, and SO2) is performed, based on Partial Mutual Information (PMI), which measures the degree of predictability of the output variable knowing the input variable, and then the daily AQI value is predicted, through PEK-based machine learning, by using the previous day’s meteorological and pollution conditions. The model is validated by a comparison between actual and predicted value, resulting in an average value of r2 = 0.58. Appl. Sci. 2018, 8, 2570 13 of 27

Meanwhile, Papaleonidas and Iliadis [47] present the spatiotemporal modelling of O3 in Athens (Greece) also using NN, speciﬁcally developing 741 ANN (21 years of data). A multi-layer feed forward and back propagation optimization are employed. The best results are conceived with three approaches—the simplest ANN, the best ANN for each case of predictions, and the dynamic variation of NN. The performance of the model is validated with the observed data and between NN approaches, resulting in r2 = 0.799, performing the best for the surrounding stations. This approach is concluded as a good option for modeling with monitoring station data due to very common gaps. Finally, Oprea et al. [48] aim to extract rules for guiding the forecasting of particulate matter (PM10) concentration levels using Reduced Error Pruning Tree (REPTree) vs. M5P (an inductive learning algorithm). The principal component analysis is used to select the best predictors, and then 27 months of eight previous-days of PM10, NO2, SO2, temperature and relative humidity data are used to predict PM10 in Romania. The validation is performed by comparing the results of two models with the observed values, concluding that M5P provides the more accurate prediction of the short-term 2 2 PM10 concentrations (r = 0.81, for one day ahead and r = 0.79 for two days ahead).

3.2.2. Category 2: Image-Based Monitoring and Tackling Low Spatial Resolution from Non-Specific/Low Resolution Sensors This group is composed of a set of seven papers. All of them address estimation issues only. It is quite logical that no forecasting studies are included in this category, because the principal objective of this kind of work is to increase the spatial resolution of the current pollution spreading. To do so, a first paper presents a method to improve the spatial resolution and accuracy of satellite-derived ground-level PM2.5 by adding a geostatistical model (Random Forest-based Regression Kriging—RFRK) to the traditional geophysical model (i.e., Chemical Transport Models—CTM) [49]. This is a two-step procedure in which a Random Forest regression models the nonlinear relationship between PM2.5 and the geographic variables (step 1) then the kriging is applied to estimate the residuals (or error) of step 1 (step 2). This work is carried out in the USA and is based on a 14 years dataset. The predictive features include CTM, satellite-derived dataset (meteorology and emissions), geographic variables (brightness of Night Time Lights—NTL, Normalized Difference Vegetation Index—NDVI, and elevation), and in situ PM2.5. The accuracy is evaluated by comparing the RFRK to CTM-derived PM2.5 models and using ground-based PM2.5 monitor measurements as reference. The results show that the RFRK significantly outperforms the other models. In addition, this method has a relatively low computational cost and high flexibility to incorporate auxiliary variables. Among the three geographic variables, elevation contributes the most and NDVI contributes the less to the prediction of PM2.5 concentrations. The main limitation of the technique is to rely on satellite images that are subject to uncertainties (e.g., data quality, completeness, and calibration). A similar approach is proposed by Just et al. [50]. The motivation of this study is to enhance satellite-derived Aerosol Optical Depth (AOD) prediction, by applying a correction based on three different Ensemble Learning algorithms: Random Forest, Gradient Boosting, and Extreme Gradient Boosting (XGBoost) The correction is implemented by considering additional inputs as predictive factors (e.g., land use and meteorology) than AOD only. This technique is applied to predict the concentration of PM2.5 in Northeastern USA during a period of 14 years. The accuracy of the method is assessed by comparing the coefficient of determination (predicted value/actual value) between the three models. The result shows that XGBoost outperforms slightly the other two algorithms. In addition, the study demonstrates that including land use and meteorological parameters in the algorithms improves significantly the accuracy when compared to the raw AOD. Nevertheless, a limitation of the technique is the fact that it requires lots of features (total number = 52). Furthermore, Zhan et al. [51] propose an improved version of the Gradient Boosting algorithm by considering the spatial nonlinearity between PM2.5 and AOD and meteorology. They develop a Geographically-Weighted Gradient Boosting Machine (GW-GBM) based on spatial smoothing kernels (i.e., gaussian) to weight the optimized loss function. This model is applied to all China and uses data Appl. Sci. 2018, 8, 2570 14 of 27 of 2014. The GW-GBM provides a better coefficient of determination than the traditional GBM. The study shows that the best features in descending order are: day of year, AOD, pressure, temperature, wind direction, relative humidity, solar radiation, and precipitation. Xu et al. [52] estimate ozone profile shapes from satellite-derived data. They develop a NN-based algorithm, called the Physics Inverse Learning Machine. The proposed method is composed of five steps. Step 1 is a clustering that gets groups of ozone profiles according to their similarity (k-means clustering procedure). Step 2 generates simulated satellite ultraviolet spectral absorption (UV spectra) of representative ozone profiles from each cluster. Step 3 consists of improving the classification effectiveness by reducing the input data through a Principal Component Analysis. In step 4 the classification model is applied for assigning an ozone profile class corresponding to a given UV spectrum. And in step 5 the ozone profile shape is scaled according to the total ozone columns. The algorithm is tested by comparing the predicted value to the observed value. 11 clusters are obtained, and the estimation error is lower than 10%. This technique can be considered an encouraging approach to predict ozone shapes based on a classification framework rather than the conventional inversion methods, even if its computational cost is relatively high. A slightly different approach is proposed by de Hoogh et al. [53]. This work aims to build global (1 km × 1 km) and local (100 m × 100 m) models to predict PM2.5 from AOD and PM2.5/PM10 ratio. It is based on an SVM algorithm to predict the concentration of PM2.5 in Switzerland. The dataset is composed of 11 years of observations and a broad spectrum of features, such as: Planetary boundary layers, meteorological factors, sources of pollution, AOD, elevation, and land use. The results show that the method is able to predict PM2.5 by using data provided by sparse monitoring stations. Another technique consists of supplementing sparse accurate station monitoring by lower fidelity dense mobile sensors, with the aim of getting a fine spatial granularity [54]. Seven Regression models are used to predict CO concentrations in Sidney, Australia. The study is divided into three stages. First, the models are built from seven years of historical data of static monitoring (15 stations) and three years of mobile monitoring. Second, the performances of each algorithm are compared. And third, field trials are conducted to validate the models. The results show that the best predictions are provided by Support Vector Regression (SVR), Decision Tree Regression (DTR), and Random Forest Regression (RFR). The validations on the field suggest that SVR has the highest spatial resolution estimation and indicates boundaries of polluted area better than the other regression models. In addition, a web application is implemented to collect data from static and mobile sensors, and to inform users. Finally, a quite different study uses image processing to predict pollution from a video-camera analysis [55]. Again, several families of machine learning algorithms are tested and compared. The method involves four sequential steps, which are: (i) camera records, (ii) image processing, (iii) feature extraction, and (iv) machine learning classification. The objective is to assess the severity of the pollution cloud emitted by factories from 12 features characterizing the image, such as— luminosity, surface of the cloud, color, and duration of the cloud. Although all the algorithms provide a similar performance, Decision Tree is the only one to be systematically classified among the models with the best (i) robustness of the parameter setting (= easy to configure), (ii) robustness to the size of the learning set, and (iii) computational cost. This study introduces interesting performance metrics based on ‘efficiency’ (higher weight for the correct classification of ‘critical events’ than ‘noncritical events’) instead of ‘accuracy’ (same weight for any event). In addition, with the objective to tackle the issue of unbalanced classes and since the authors are more interested in critical events (black cloud), which are less represented, some ‘noncritical events’ are removed from the learning set in order to get a balanced representation of the classes. Nevertheless, the application of this method is limited to the assessment of the contamination produced by plants. Appl. Sci. 2018, 8, 2570 15 of 27

3.2.3. Category 3: Considering Land Use and Spatial Heterogeneity/Dependence Four out of five papers classified in this category are estimation models. The most general manuscript is written by Abu Awad et al. [56]. The objective is to build a land use model to predict Black Carbon (PM10) concentrations. The approach is divided into two steps. First, a land use regression model is improved by using a nu-Support Vector Regression (nu-SVR). Second, a generalized additive model is used to refit residuals from nu-SVR. This study is applied in New England States (USA) and exploits a 12 years dataset, which stores data from 368 monitoring stations. A broad spectrum of variables is used as features, such as: Filter analysis, spatial predictors (elevation, population density, traffic density, etc.), and temporal predictors (meteorology, planet boundary layer, etc.). The model is tested in cold and warm seasons and is assessed by comparison to actual data. Overall, the results provide a high coefficient of determination (r2 = 0.8). In addition, it appears that this coefficient is significantly higher in the cold than in the warm season. Araki et al. [57] also build a spatiotemporal model based on land use to capture interactions and non-linear relationships between pollutants and land. The study consists of comparing two algorithms—Land Use Random Forest (LURF) vs. Land Use Regression (LUR)—to estimate the concentration levels of NO2. This work takes place in the region of Amagasaki, in Japan, over a period of four years. Besides land use, the authors use population, emission intensities, meteorology, satellite-derived NO2, and time as features. The accuracy of LURF outperforms slightly LUR, and the coefficient of determination gets a same range of values as in the above study. LURF is a bit better than LUR, because the former allows for a non-linear modelling whereas the latter is limited to a linear relationship. Another advantage of using random forest is to produce an automatic selection of the most relevant features. The results show that the best predictors in descending order are—green area ratio, satellite-based NO2, emission sources, month, highways, and meteorology. The variable population is discarded from the model. Nevertheless, the interpretation of the regression model is easier than the random forest-based model, because the former provides coefficients representing the direction and magnitude of the effects of predictor variables. The variable selection issue related to LUR is tackled in the study carried out by Beckerman et al. [58]. They develop a hybrid model that mix LUR with deletion, substitution, and addition machine learning. Thanks to this method, no more than 12 variables are used to predict the concentration of PM2.5 and NO2. The case study is applied to all California and uses data on the period 1999–2002 for PM2.5, and 1988–2002 for NO2. The overall performance of the model is fair, but the accuracy for estimating PM2.5 is significantly lower than NO2. Brokamp et al. [31] perform a deeper analysis based on a prediction of the concertation of the chemical composition of PM2.5. As Beckerman et al. [57], they compare the performance of a LURF versus LUR approach. The dataset is built from the measurements of 24 monitoring stations in the city of Cincinnati (USA) during the period 2001–2005. Over 50 spatial parameters are registered, which include transportation, physical features, socioeconomic characteristics, greenspace, land cover, and emission sources. Again, LURF outperforms slightly LUR. The originality of this work is to create models that predict not only the air quality, but also the individual concentration of metal components in the atmosphere from land use parameters. Yang et al. [59] is the only study of this category that provides a forecasting model. The approach is applied in three steps. First, a clustering analysis is performed to handle the spatial heterogeneity of PM2.5 concentrations. Several clusters are defined, based on homogeneous subareas of the city of Beijing (China). Second, input spatial features that address the spatial dependence are calculated using a Gauss vector weight function. And third, spatial dependence variables and meteorological variables are used as input features of an SVR-based model of prediction. The performance of the proposed Space-Time Support Vector Regression (STSVR) algorithm is compared to the ARIMA, traditional SVR, and NN. The best model depends on the forecasting time span. STSVR outperforms the prediction of the other models from 1 h to 12 h ahead, whereas the global SVR outperforms STSVR from 13 h to 24 h ahead. This study demonstrates that (i) the relationship between air pollutant concentrations Appl. Sci. 2018, 8, 2570 16 of 27 and other relevant variables changes over spatial areas, and (ii) as the forecasting time increases, the prediction accuracy decreases more with STSVR than SVR.

3.2.4. Category 4: Hybrid Models and Extreme/Deep Learning This set of papers represents one of the rare categories in which forecasting models (10) are significantly more frequent than estimation models (2). The first estimation study consists of building a reliable model to predict NO2 and NOx with a high spatiotemporal resolution and over an extended period of time [60]. This work is carried out in three steps. First, multiple non-linear (traffic, meteorology, etc.), fixed (e.g., population density) and spatial (e.g., elevation) effects are incorporated to characterize spatiotemporal variability of NO2 and NOx concentration. Second, the Ensemble Learning algorithm is applied to reduce variance and uncertainty in the prediction based on step 1. Third, an optimization process is implemented by tackling possible incomplete time-varying covariates recording (traffic, meteorology, etc.) in order to get a continuous time series dataset. The prediction performance is quite high (r2 ≈ 0.85). Traffic density, population density, and meteorology (wind speed and temperature) account for 9–13%, 5–11%, and 7–8% of the variance, respectively. Nevertheless, the method is less accurate for the prediction of low pollution levels. A different estimation approach is proposed by Zhang and Ding [61]. The authors use an Extreme Learning Machine (ELM) to tackle the low convergence rate and the local minimum that characterize the NN algorithms. Their ELM consists of only 2-layer NN. The first layer (hidden layer) is fixed and random (its weight does not need to be adjusted). The second layer only is trained. They estimate a broad spectrum of contaminants (NO2, NOx,O3, PM2.5, and SO2) from meteorological and time parameters. It appears that the ELM is significantly better than NN and Multiple Linear Regression (MLR), because it provides a higher accuracy and a lower computational cost. An even more advanced study is carried by Peng et al. [62]. The authors propose an efficient non-linear machine learning algorithm for air quality forecasting (up to 48 h), which is easily updatable in ‘real time’ thanks to a linear solution applied to the new data. Five algorithms are compared: MLR, Multi-Layer Neural Network (MLNN), ELM, updated-MLR, and updated-ELM. This approach involves two steps and predict three types of contaminants (O3, PM2.5, and NO2). Step 1 consists of the initial training stage of the algorithm. The 2nd step involves a sequential learning phase, in order to proceed with an online (daily) update of the MLR and ELM, and only seasonally (trimonthly) update of the MLNN. The results show that the updated-ELM tends to outperform the other models, in terms of R2 and MAE. In addition, the linear updating applied to MLR and ELM is less costly than updating a MLNN. Nevertheless, all the models tend to underpredict extreme values (the non-linear models even more than the linear models). Another type of forecasting involves hybrid models. This is the case of the study performed by Zhang et al. [63], who predict short and long-term CO from a hybrid model composed with Partial Least Square (PLS), for data selection, and SVM, for modelling. Overall, SVM-PLS seems more performant than traditional SVM, because both the precision and computational cost of the former is better than the latter. However, the accuracy is lower for hourly than daily forecasting. Tamas et al. [64] propose a different hybrid model to forecast air contaminants (O3, NO2, and PM10), with a special focus on the critical prediction of pollution peaks. The hybrid model is composed with NN and clustering and it is compared to the Multilayer Perceptron. The proposed algorithm outperforms the traditional NN, especially in the prediction of PM10 and O3 peaks of pollution. Since performing a forecasting consists mainly of predicting a time series, Ni et al. [65] propose a hybrid model that implements a NN and an Autoregressive Integrated Moving Average (ARIMA). The concentration of PM2.5 are predicted in Beijing (China) from meteorological parameters, chemical variables, and microblog data. Such an approach allows for a good forecast for a few hours ahead, but the bigger the time lag, the bigger the error. Appl. Sci. 2018, 8, 2570 17 of 27

A time series analysis is also used by Li and Zhu [66] to predict the AQI (NO2 + PM10 + O3 + PM2.5) 1 h ahead and from the extraction of conceptual features (randomness and fuzziness of time series data). A hybrid model based on a technique of Cloud Model Granulation (CMG) and SVR is implemented in two steps. First, CMG uses probability and fuzzy theory to extract from time series the concept features of: randomness and fuzziness of the data. And second, after extracting these features, an SVR is used for the prediction per se. The performance of CMG-SVR is slightly better than the individual application of the same techniques (CMG and SVR alone) or a NN. Nevertheless, this method is still only able to make a short-term prediction (about 1 h ahead). The fuzzy approach to forecast PM10 hourly concentration 1 h ahead is also used by Eldakhly et al. [67]. To tackle the randomness and fuzziness of the data, the hybrid model applies, first, a chance weight value to the target variable in order to minimize an outliner point (if any) that can be used, afterward, as a support vector point during the training process. The result of this study is one of the rare examples that shows a model that outperforms ensemble learning algorithms (Boosting and Stacking). Wang et al. [68] propose to optimize the AQI forecasting by applying a ‘decomposition and ensemble’ technique in three steps: (i) decomposition of the original AQI time series into a set of independent components (i.e., frequencies or IMFs), (ii) prediction of each component (by ELM), and (iii) aggregation of the forecast values of all components. The hybrid model takes every eight successive days as AQI data inputs of the ELM to forecast the ninth day. The originality of the study is to propose a decomposition of the AQI time series into two phases— by Complementary Ensemble Empirical Mode Decomposition (CEEMD), for low and medium frequencies, and by Variation Mode Decomposition (VMD), for high frequencies. This 2-phase decomposition provides a superior forecasting accuracy than 1-phase decomposition implementations. Fuzzy logic, ELM, and heuristic are put together in the model developed by Li et al. [66]. This study forecasts the AQI (based on PM10, PM2.5, SO2, NO2, CO, and O3) of six cities in China. Fuzzy logic is used as a feature selection, because relevant predictive contaminants are not the same according to the city. ELM and heuristic optimization algorithms are applied for a deterministic prediction of the contaminant concentrations. This hybrid model gives a better prediction than other NN and time series (e.g., ARIMA) algorithms but it is slightly slower than the NNs. More recently, a bunch of studies have implemented a Deep Learning approach to improve the accuracy of air pollution forecasting. For instance, Zhao et al. [69] propose a Deep Recurrent Neural Network (DRNN) method to forecast daily Air Quality Classification (AQC). This technique is divided into two steps. First, data of six pollutants are pre-processed, in order to group the concentrations into four categories or Individual Air Quality Index—IAQI (from unhealthy to good). Second, an RNN based on Long Short-Term Memory (LSTM) is used to perform the forecasting. LSTM is an RNN, which includes a memory that permits to learn the input sequence with longer time steps (e.g., problems related to time series). Despite the high expectations, the predictive accuracy of the model is not significantly better than the performance of the other tested algorithms (SVM and Ensemble Learning). In addition, the disadvantage of Deep Learning is a high computational cost and a low interpretability of the model. A more satisfactory result is obtained by Huang and Kuo [70], who also use a hybrid model based on NN and LSTM to forecast PM2.5 concentrations 1 h ahead, by processing big data. The authors use a Convolutional Neural Network (CNN) instead of the traditional NN, because its partially connected architecture allows for a reduced training time. This optimized model seems to outperform all the principal machine learning algorithms (SVM, Random Forest, and Multilayer Perceptron). However, the relevance of such a technique still has to be confirmed over a longer time span forecast.

3.2.5. Category 5: Towards an Application System With only four papers out of 46, the application category is one of the smallest. This result means that the majority of the published studies are still very theoretical. A first interesting application system is proposed by Sadiq et al. [71]. The purpose of their model is to regulate the traffic, in order to Appl. Sci. 2018, 8, 2570 18 of 27 optimize pollution level while maximize the vehicle flow. The method is based on a hybrid model that involves problem-solving, multi-agent system, and NN to estimate the concentration of O3 in Marrakech-City (Morocco). The predictive features are traffic, pollutant covariates, and meteorological parameters (relative humidity, wind speed, temperature, and solar radiation). The accuracy of the model is assessed by comparing the predicted values to the measurements at the monitoring stations. The system is optimized by using the Hadoop framework that manages big data and a multi-agent software architecture. The objective is to provide the drivers with recommendations on the best path (sum of the shortest + least polluted road) after applying a search graph algorithm (Dijkstra). An app of the system is currently available. Other studies intend to provide a public application to forecast air quality. This is the case of the work carried out by Tzima et al. [72], who develop a tool that is integratable with existing Environmental Management Systems implemented in Greece. The machine learning model implements Decision Rules to forecast pollution peaks of O3 and PM10. The obtained model is a trade-off between performance and understandability. It is less accurate than a linear regression to forecast PM10 but it is better than the other tested models to predict O3. A more advanced system is based on the operational forecasting platform Prev’Air [73]. This platform aims at forecasting maps, on a daily basis, for O3, NO2, and PM10. This study implements a Ridge Regression method that reduces the errors of the forecasts (RMSE) for all the contaminants (by 35% for O3, 26% for NO2, and 19% for PM10). However, the technique presents some limitations to predict pollution peaks. In that sense, it is not adapted for cities with frequent violations of air quality standards. The last work consists of a network installation of low-cost pollution sensors [74]. The whole system is able to store data, to process these raw data for a posterior forecasting, and to present the prediction through different channels, such as mobile application and Web portal. Different machine learning methods (SVM, Regression, Model Trees, and NN) are tested, in order to identify the best algorithm to forecast the concentration levels of O3, NO2, and SO2. It seems that the Model Trees provide the lowest RMSE. This solution is particularly relevant to tackle low density monitoring stations and could be advantageously applied in developing countries that suffer high pollution rates. Overall, the majority of the application systems implement traditional machine learning techniques.

3.2.6. Category 6: Nanoparticles as a New Challenge In the last category, the studies focus on estimating the concentrations and peaks of nanoparticles. A number of years of research indicates that the levels of PM2.5 concentrations are dropping in most of the high- and mid-income countries [75–78]. However, in the case of nanoparticles, the harm comes not from their mass (particle concentration, usually high for the larger size aerosols), which compared to micro-sized particles is minute, but the quantity (particle number, usually very high for the smaller size aerosols) entering the human blood stream and damaging the cardiovascular system. Recent study indicates that nanoparticles attack the injured blood vessels and may cause heart attacks and strokes, while long-term exposure causes vascular damage [79]. Therefore, studying these ultrafine particles (UFP) is quite important due to the increased production of very small particles by modern engines. However, not many cities have the infrastructure to monitor these complex pollutants. Therefore, Al–Dabbous et al. [80] aim to build a model that estimates the concentration of nanoparticles from accessible and routinely-monitored pollution and meteorological factors using a feed-forward Neural Network with back-propagation. First, different architectures of NN (number of layers and neurons) are tested through an empiric approach (i.e., trial and error), in order to select the models that provide the minimum error of prediction according to the features considered, then the seven best models (each one with different types and number of features) are compared, and finally, in order to assess the prediction of the models for high-pollution concentrations 75th percentile values are considered. One month of five pollutant covariates (PM10, SO2,O3, NOx, and CO), and/or meteorological data (wind speed, and temperature) are used in Fahaheel City (Kuwait). The model performance is assessed by comparison between the seven models (i.e., type and number Appl. Sci. 2018, 8, 2570 19 of 27 of features considered) based on the determination coefficient. The models that include meteorology have the best results r2 = 0.8 (with meteorology), and r2 = 0.74 (without meteorology). The accuracy of the models is slightly lower for extreme concentrations (>75th percentile values). The study shows that the concentration in nanoparticles is very sensitive to meteorological factors. Similarly, Pandey et al. [81] aim to understand the relationships between the concentration of submicron particles and meteorological and traffic factors in order to estimate nanoparticles (particulate matter less than 1 µm—PM1, and UFP) in Hangzhou (China). As in the previous study [80], a short term, three days in winter data of meteorological (temperature, relative humidity, wind speed and direction, precipitation and pressure), traffic (flow, speed) and time parameters are used. Tree-based classification models, SVM, Naïve Bayes, Bayesian Network, NN, k-NN, Ripper and RF are implemented. Correlation with PM and each predictor is performed, to know which single feature has better predicting power, and based on that the multiple parameter model is used. Data are split into two (low vs. high) and three (low vs. medium vs. high) ranges of pollution classes. Model performance is validated against the observed data (precision, area under curve—AUC) resulting in tree algorithm AUC and f-measure 1 for PM1 (ADTree, RF), and for UFP 0.84 (RF). This study is a promising example in predicting nanoparticles without actual measurements. It shows that it is easier to predict PM1, than UFP. Tree algorithms are able to produce almost completely 2 accurate prediction of PM1, and very good (r = 0.844) for UFP. A binary split is better than ternary for all classifiers. Finally, weather also shows a strong relationship with PM1, while traffic parameter is important for UFP.

3.3. Synthesis In order to synthetize the results of this systematic review, an MCA is performed. This factorial analysis looks for the principal dimension explaining the variability between the papers and to closely examine relationships between them. The MCA dimensions are quantitative variables which summarize the qualitative variables that characterize each study. This statistical method suggests links between these qualitative variables by the observation of their proximity in the point cloud-based representation of the categories (Figure5). Here, the total percentage of inertia is 28%, which is an acceptable value considering the fact the papers are located in a high K–J dimensional space. The x-axis (F1) shows an opposition between forecasting models and estimation models. On the one hand, the forecasting models seem more used to predict Air Quality Index (prediction-AQI) and are mainly based on NN (algorithm-NN) and SVM (algorithm-SVM) algorithms. They do not rely on land use predictors (land-F), but a lot on the targeted predicted parameter (target-T). Since a forecasting usually involves a time series analysis, it makes sense to use the targeted parameter as predictive feature for predicting its evolution over time. Considering the large distance between ‘model-forecasting’ and ‘prediction-NOx’, it is also possible to conclude that this kind of modelling is not used to predict the nitrogen oxides in the atmosphere. On the other hand, the estimation models constitute the main approach to predict ozone (prediction-O3) and nanoparticles (prediction-PM01) and are mostly based on Ensemble Learning algorithms (algorithm-Ensemble). Contrary to forecasting, they do not use the predicted factor as feature (target-F), because this type of model focuses on estimating the concentration of a determined pollutant from other sources of contamination or dispersion. This aspect explains the reason why atmospheric parameters (atmosphere-T), time (time-T), satellite images (image-T), and land use (land-T) tend to be located on the same side as ‘model-estimation’. The y-axis (F2) explains less variables than the x-axis (F1). Nevertheless, it highlights a clear opposition between 2 peculiars algorithms that appear only in 2 different papers (algorithm-Rules and algorithm-Lazy). If we look at the data, it is possible to conﬁrm that these two studies address distinct types of modelling (estimation vs. forecasting) and predicted parameters (PM10 vs. AQI). Appl. Sci. 2018, 8, x FOR PEER REVIEW 19 of 27 hand, the forecasting models seem more used to predict Air Quality Index (prediction-AQI) and are mainly based on NN (algorithm-NN) and SVM (algorithm-SVM) algorithms. They do not rely on land use predictors (land-F), but a lot on the targeted predicted parameter (target-T). Since a forecasting usually involves a time series analysis, it makes sense to use the targeted parameter as predictive feature for predicting its evolution over time. Considering the large distance between ‘model-forecasting’ and ‘prediction-NOx’, it is also possible to conclude that this kind of modelling is not used to predict the nitrogen oxides in the atmosphere. On the other hand, the estimation models constitute the main approach to predict ozone (prediction-O3) and nanoparticles (prediction-PM01) and are mostly based on Ensemble Learning algorithms (algorithm-Ensemble). Contrary to forecasting, they do not use the predicted factor as feature (target-F), because this type of model focuses on estimating the concentration of a determined pollutant from other sources of contamination or dispersion. This aspect explains the reason why atmospheric parameters (atmosphere-T), time (time-T), satellite images (image-T), and land use (land-T) tend to be located on the same side as ‘model-estimation’. The y-axis (F2) explains less variables than the x-axis (F1). Nevertheless, it highlights a clear opposition between 2 peculiars algorithms that appear only in 2 different papers (algorithm-Rules and algorithm-Lazy). If we look at the data, it is possible to confirm Appl.that Sci.these2018 two, 8, 2570 studies address distinct types of modelling (estimation vs. forecasting) and predicted20 of 27 parameters (PM10 vs. AQI).

Figure 5. MCA representation representation of the point cloud of of the the ca categoriestegories that that characterize characterize the the selected selected papers. papers. The studies are described from nine variables (J) and 27 categories (K). Variable 1 is the type of modelling and has twotwo categories:categories: Estimation (model-estimation)(model-estimation) and forecast (model-forecasting). Variable 22 isis thethe typetype ofof algorithmalgorithm andand hashas sixsix categories:categories: EnsembleEnsemble LearningLearning (algorithm-Ensemble), Neural Network (algorithm-NN), Support Vector MachineMachine (algorithm-SVM),(algorithm-SVM), MultipleMultiple RegressionRegression (algorithm-Regression),(algorithm-Regression), Decision Rules (algorithm-Rules),(algorithm-Rules), and Lazy methodsmethods (algorithm-Lazy). Variable 33 isis thethe predictedpredicted parameterparameter andand hashas sevenseven categories:categories: Air Quality Index (prediction-AQI),

PM1010 (prediction-PM10),(prediction-PM10), PM 2.5 (prediction-PM25),(prediction-PM25), nanoparticles nanoparticles (p (prediction-PM01),rediction-PM01), carbon monoxide (prediction-CO),(prediction-CO), nitrogennitrogen oxidesoxides (prediction-NOx),(prediction-NOx), andand ozoneozone (prediction-O3).(prediction-O3). Variable Variable 4 is thethe feature target and has two categories: Used (target-T) and not used (target-F). Variable 5 is the feature pollutant covariates and has two categories: Used (covariates-T) and not used (covariates-F). Variable 6 is the feature atmospheric parameters and has two categories: Used (atmosphere-T) and not used (atmosphere-F). Variable 7 is the feature land use and has two categories: Used (land-T) and not used (land-F). Variable 8 is the feature time and has two categories: Used (time-T) and not used (time-F). Variable 9 is the feature image processing and has two categories: Used (image-T) and not used (image-F).

feature target and has two categories: Used (target-T) and not used (target-F). Variable 5 is the feature pollutant covariates and has two categories: Used (covariates-T) and not used (covariates-F). Variable 6 is the feature atmospheric parameters and has two categories: Used (atmosphere-T) and not used (atmosphere-F). Variable 7 is the feature land use and has two categories: Used (land-T) and not used (land-F). Variable 8 is the feature time and has two categories: Used (time-T) and not used (time-F). Variable 9 is the feature image processing and has two categories: Used (image-T) and not used (image-F).

Overall, more studies are dedicated to estimation (28) than forecasting (18). There are only two out of six categories in which the number of forecast motivations is higher than estimation purposes (Figure 6). This is the case of Category 4, which presents hybrid solutions to predict air contamination and includes deep and extreme learning. These techniques are usually characterized by a high performance of prediction, which is particularly adapted to tackle the uncertainty to forecast concentrations of pollutant in the near or far future. The other group is Category 5, which focuses on implementing practical applications. Intuitively, it makes more sense to develop an App or a Web Appl.portal Sci. that2018 provides, 8, 2570 the users with a forecasting information that supports the planning of their21 next of 27 activities than to propose a theoretical assessment of air quality. On the contrary, the other four categories are significantly more represented by estimation than forecasting models. Obviously, the signiﬁcantly more represented by estimation than forecasting models. Obviously, the group with the group with the most estimations is Category 1, because it is interested in understanding the most estimations is Category 1, because it is interested in understanding the relationship between the relationship between the sources of contamination and air quality. According to the MCA, it also sources of contamination and air quality. According to the MCA, it also tends to prefer models that tends to prefer models that involve a high interpretability (e.g., Regression or Decision Tree-based involve a high interpretability (e.g., Regression or Decision Tree-based models) than accuracy. A similar models) than accuracy. A similar interpretation can be made for Category 4, which includes land use interpretation can be made for Category 4, which includes land use parameters and other spatial parameters and other spatial heterogeneities of the environment to increase the comprehensiveness heterogeneities of the environment to increase the comprehensiveness of the models. Interestingly, of the models. Interestingly, there are two groups (Categories 2 and 6) in which forecasting models there are two groups (Categories 2 and 6) in which forecasting models are not represented at all. are not represented at all. Only estimation models are built in Category 2, certainly because the Only estimation models are built in Category 2, certainly because the priority of this kind of study is priority of this kind of study is improving the spatial resolution of the current pollution spreading improving the spatial resolution of the current pollution spreading and not predicting it hours ahead. and not predicting it hours ahead. Finally, considering that Category 6 is a new challenge, it is natural Finally, considering that Category 6 is a new challenge, it is natural that the only two studies on the that the only two studies on the topic focus on an estimation before carrying out a forecast. The topic focus on an estimation before carrying out a forecast. The identiﬁcation of these two unexplored identification of these two unexplored axes allows us to suggest where the new research should be axes allows us to suggest where the new research should be evolving next. evolving next.

Figure 6. Number of types of studies (estimation(estimation vs. forecasting)forecasting) per group. Category 1: Identifying relevant predictors and understandingunderstanding the non-linear relationship with air pollution.pollution. Category Category 2: 2: Image-based monitoring and tackling low spatial resolution from non-specific/low non-speciﬁc/low resolution resolution sensors. sensors. Category 3: Considering land use and spatial heterogeneity/dependence.heterogeneity/dependence. Category 4: Hybrid models and extreme/deep learning. learning. Category Category 5: 5: Towards Towards an application application system. Category 6: Nanoparticles Nanoparticles as a new challenge.

Finally, the performances of all the studied modelsmodels were evaluated in general per algorithm and then each algorithm performance per pollutant for estimation and forecasting studies. For the firstfirst analysis, we selected the most used algorithms: Ensemble, NN, SVM and Regression modeling to evaluate coefficients of determination resulting from all 46 studies. The best overall performance is registered for the Ensemble modeling (r2 = 0.79) with a low variability in the results between different studies. Second best is the Regression modeling (r2 = 0.74) with also low standard deviation. These two types of algorithms are mostly used in the estimation modeling (see Figure5), thus resulting in a better predictive performance. SVM has also showed a good performance (r2 = 0.67 ± 0.15). This algorithm is used for the forecasting and a few studies on estimation modeling. The lowest performance was registered from NN modeling (r2 = 0.64 ± 0.27), often used in forecasting studies (see Figure5). In order to further identify a success of forecasting and estimation studies we separated the resulting available coefficients of determination for each pollutant. Figure7 shows that in almost all the cases, with the exception of PM2.5 the estimation modeling results in better accuracy than the forecasting modeling. However, the variation of the results is significantly higher in the forecasting. 2 2 2 The estimation modeling performs quite well for CO (r = 0.76), NOx (r = 0.78), O3 (r = 0.74) and 2 PM2.5 (r = 0.72). Meanwhile, the forecasting models reach a very high level of performance for Appl. Sci. 2018, 8, x FOR PEER REVIEW 21 of 27 analysis, we selected the most used algorithms: Ensemble, NN, SVM and Regression modeling to evaluate coefficients of determination resulting from all 46 studies. The best overall performance is registered for the Ensemble modeling (r2 = 0.79) with a low variability in the results between different studies. Second best is the Regression modeling (r2 = 0.74) with also low standard deviation. These two types of algorithms are mostly used in the estimation modeling (see Figure 5), thus resulting in a better predictive performance. SVM has also showed a good performance (r2 = 0.67 ± 0.15). This algorithm is used for the forecasting and a few studies on estimation modeling. The lowest performance was registered from NN modeling (r2 = 0.64 ± 0.27), often used in forecasting studies (see Figure 5). In order to further identify a success of forecasting and estimation studies we separated the resulting available coefficients of determination for each pollutant. Figure 7 shows that in almost all the cases, with the exception of PM2.5 the estimation modeling results in better accuracy than the forecastingAppl. Sci. 2018 modeling., 8, 2570 However, the variation of the results is significantly higher in the forecasting.22 of 27 The estimation modeling performs quite well for CO (r2 = 0.76), NOx (r2 = 0.78), O3 (r2 = 0.74) and PM2.5 2 2 (r = 0.72).2 Meanwhile, the forecasting models reach a very high level of performance for PM2.5 (r = PM2.5 (r = 0.74), which is an important achievement, as CTMs often struggle with this specific 0.74), which is an important achievement, as CTMs often struggle with this specific pollutant [49,82]. pollutant [49,82]. Interestingly, the worst performance of forecasting is registered for the traffic-related Interestingly, the worst performance of forecasting is registered for the traffic-related pollutants CO pollutants CO and NOx. and NOx.

2 FigureFigure 7. 7. ModelModel performance performance evaluated evaluated with with the the averag averagee and and standard standard deviation deviation of of the available r 2 perper pollutant pollutant of of estimation estimation and and forecast forecast studies.

4.4. Conclusions Conclusions ThisThis manuscript manuscript examines 4646 selectedselected scientificscientific articles articles focusing focusing on on machine machine learning learning approach approach to topredict predict air air pollution. pollution. While While machine machine learning learning application application inin atmosphericatmospheric sciences has been been rapidly rapidly increasingincreasing in in the the last last two two years, years, this this growth growth ha hass been been restricted restricted to to Eurasian Eurasian and and North North American American continents,continents, with with only only four four studies studies in in the the southern southern hemisphere. hemisphere. ThisThis systematic systematic review review shows shows that that there there are are two two kinds kinds of of studies studies that that implement implement a a machine machine learninglearning approach approach for for predicting predicting air air quality. quality. The The fi firstrst class, class, the the majority, majority, is is represented represented by by papers papers thatthat focus focus on on an an estimation estimation of of the the concentration concentration of of pollutants. pollutants. It mostly It mostly uses uses Ensemble Ensemble Learning Learning or Regressionor Regression algorithms, algorithms, because because they theyprovide provide an excellent an excellent trade-off trade-off between between interpretability interpretability and performanceand performance of the ofmodel. the model.The second The class second is co classmposed is composedof manuscripts of manuscripts that address that forecasting address problems.forecasting This problems. kind of Thisquestion kind is of clearly question treated is clearly by the treated use of byNN the and use SVM of NN techniques. and SVM The techniques. forecast prioritizesThe forecast the prioritizes accuracy theover accuracy the interpretability, over the interpretability, which may explain which maywhy explainsuch powerful why such algorithms, powerful butalgorithms, also considered but also consideredas a black asbox, a black are preferre box, ared. preferred. Our synthesis Our synthesis shows showsthat the that precision the precision of the of forecastingthe forecasting tends tends to be to lower be lower and and more more variable variable than than the the precision precision of of the the estimation. estimation. This This fact fact may may justifyjustify the the necessity necessity to to apply apply methods methods that that are are computationally computationally more more demanding demanding (e.g., (e.g., deep deep learning) learning) toto tackle tackle the the complexity complexity to to predict predict the the value value of of contaminants contaminants hours or days ahead. Overall, it seems that machine learning is an appropriate method for the prediction of air pollution. The best example is the case of fine particulate matter. Estimating such contaminants is particularly challenging for the traditional CTM approach. With machine learning, the accuracy of both estimation and forecasting of PM2.5 reaches one of the highest values if compared with the other pollutants. Nevertheless, the models still have reduced precision for the prediction of extreme concentrations of atmospheric pollutants. In general, the accuracy for the prediction of high pollution peaks is lower than for medium and small peaks. Also, the forecasting performance is still limited for certain chemicals, such as CO and NOx. In addition, the models seem to perform better for more extreme weather conditions (e.g., fall, winter, windier, etc.). Finally, it is important to note that an increasing and dangerous form of contamination by nanoparticulate is understudied (only two papers identified in this selection). Thus, we consider that improving the models that predict pollution peaks and critical pollutants constitute the next challenges that should be addressed by future machine Appl. Sci. 2018, 8, x FOR PEER REVIEW 22 of 27

Overall, it seems that machine learning is an appropriate method for the prediction of air pollution. The best example is the case of fine particulate matter. Estimating such contaminants is particularly challenging for the traditional CTM approach. With machine learning, the accuracy of both estimation and forecasting of PM2.5 reaches one of the highest values if compared with the other pollutants. Nevertheless, the models still have reduced precision for the prediction of extreme concentrations of atmospheric pollutants. In general, the accuracy for the prediction of high pollution peaks is lower than for medium and small peaks. Also, the forecasting performance is still limited for certain chemicals, such as CO and NOx. In addition, the models seem to perform better for more Appl. Sci.extreme2018, 8weather, 2570 conditions (e.g., fall, winter, windier, etc.). Finally, it is important to note that an23 of 27 increasing and dangerous form of contamination by nanoparticulate is understudied (only two papers identified in this selection). Thus, we consider that improving the models that predict learningpollution algorithms. peaks and This critical conclusion pollutants should constitute be moderated the next bychallenges the limitations that should of this be systematicaddressed by review, whichfuture was machine restricted learning to studies algorithms. published This inconclusion scientific should journals be moderated only. by the limitations of this systematic review, which was restricted to studies published in scientific journals only. Author Contributions: Conceptualization, Y.R. and R.Z.; methodology, Y.R.; formal analysis, Y.R. and R.Z.; data curation,Author Contributions: Y.R. and R.Z.; Conceptualization, data analysis, Y.R. Y.R. and and R.Z.;R.Z.; methodology, writing and editing,Y.R.; formal Y.R. analysis, and R.Z. Y.R. and R.Z.; data Funding:curation,This Y.R. research and R.Z.; received data analysis, no external Y.R. and funding. R.Z.; writing and editing, Y.R. and R.Z. ConflictsFunding: of Interest: This researchThe authors receiveddeclare no external no conflictfunding. of interest. Conflicts of Interest: The authors declare no conflict of interest. Appendix A Appendix A,

REFERENCE group model algorithm prediction target covariates atmospherelandtimeimage Lin et al., 2018 hybrid forecasting SVM AQI T F F F F F Araki et al., 2018 land use estimation Ensemble NOx T F T T T T Huang & Kuo, 2018 hybrid forecasting NN PM25 T F T F F F Li & Zhu, 2018 hybrid forecasting NN AQI T F F F F F Grange et al., 2018 predictors estimation Ensemble PM10 F F T F T F Just et al., 2018 spatial resolution estimation Ensemble PM25 F F T T F T Yang et al., 2018 land use forecasting SVM PM25 T F T F F F Liu et al., 2018 spatial resolution estimation Ensemble PM25 T F F F F T Eldakhly et al., 2018 hybrid forecasting SVM PM10 T F T F T F Zhao et al., 2018 hybrid forecasting NN AQI T F F F F F Zhan et al., 2018 predictors estimation Ensemble O3 F F T T T F De Hoogh et al., 2018 spatial resolution estimation SVM PM25 F F T T F T Chen et al., 2018 predictors forecasting NN AQI T F T F F F Martínez-España et al., 2018 predictors estimation Ensemble O3 F T T F F F Xu et al., 2017 spatial resolution estimation NN O3 F F F F F T Li et al., 2017 hybrid estimation Ensemble NOx F F T T F F Liu et al., 2017 predictors estimation SVM AQI T F T F F F Kleine Deters et al., 2017 predictors estimation Regression PM25 F F T F T F Hu et al., 2017 spatial resolution estimation SVM CO T F F F T F Al-Dabbous et al., 2017 nanoparticles estimation NN PM01 F T T F F F Peng et al., 2016 hybrid forecasting NN AQI T F T F F F Wang et al., 2017 hybrid forecasting NN AQI T F F F F F Zhang & Ding, 2017 hybrid estimation NN AQI F F T F T F Ni et al., 2017 hybrid forecasting NN PM25 T T T F F F Brokamp et al., 2017 land use estimation Ensemble PM25 F F T T F F Abu Awad et al., 2017 land use estimation SVM PM10 F F T T F F Zhan et al., 2017 spatial resolution estimation Ensemble PM25 T F T F F T Sadiq et al., 2016 application estimation NN O3 F T T T F F Carnevale et al., 2016 predictors estimation Lazy PM10 T T F F F F Bougoudis et al., 2016 predictors estimation Ensemble AQI T T F F F F Shaban et al., 2016 application forecasting Ensemble AQI F T T T T F Tamas et al., 2016 hybrid forecasting NN AQI T F T F F F Oprea et al., 2016 predictors forecasting Ensemble PM10 T T T F F F Nieto et al., 2015 predictors estimation Regression AQI T F F F F F Sayegh et al., 2014 predictors estimation Regression PM10 T T F F F F Debry & Mallet, 2014 application forecasting Regression AQI T F F F F F Singh et al., 2013 predictors estimation Ensemble AQI T F T F F F Papaleonidas & Iliadis, 2013 predictors forecasting NN O3 T T F F F F Beckerman et al., 2013 land use estimation Regression NOx F T F T F T Pandey et al., 2013 nanoparticles estimation Ensemble PM01 F F T T T F Philibert et al., 2013 predictors estimation Ensemble NOx F T F F F F Vong et al., 2012 predictors forecasting SVM AQI T F T F F F Yeganeh et al., 2012 hybrid forecasting SVM CO F T F F F F Suárez Sánchez et al., 2011 predictors estimation SVM AQI T F F F F F Gacquer et al., 2011 spatial resolution estimation Ensemble AQI F F F F F T Tzima et al., 2011 application forecasting Rules AQI T F T F F F

Figure A1. Source table used to perform the Multiple Correspondence Analysis (MCA).

References

1. WHO. 7 Million Premature Deaths Annually Linked to Air Pollution Page 1 of 2 WHO|7 Million Premature Deaths Annually Linked to Air Pollution Page 2 of 2. Available online: http://www.who.int/mediacentre/ news/releases/2014/air-pollution/en/#.WqBfue47NRQ.mendeley (accessed on 7 March 2018). 2. Limb, M. Half of wealthy and 98% of poorer cities breach air quality guidelines. BMJ 2016, 353.[CrossRef] [PubMed] 3. WHO. Air Pollution Levels Rising in Many of the World’s Poorest Cities. Available online: http://www.who. int/mediacentre/news/releases/2016/air-pollution-rising/en/#.WhOPG9ANlBk.mendeley (accessed on 21 November 2017). 4. Daly, A.; Zannetti, P. Air pollution modeling—An overview. Ambient Air Pollut. 2007, I, 15–28. Appl. Sci. 2018, 8, 2570 24 of 27

5. Met Office. Numerical Atmospheric-Dispersion Modelling Environment (NAME) Model. Available online: http://www-cast.ch.cam.ac.uk/cast_pics/WP_NAME.pdf (accessed on 5 December 2018). 6. Cohen, A.J.; Brauer, M.; Burnett, R.; Anderson, H.R.; Frostad, J.; Estep, K.; Balakrishnan, K.; Brunekreef, B.; Dandona, L.; Dandona, R.; et al. Estimates and 25-year trends of the global burden of disease attributable to ambient air pollution: An analysis of data from the Global Burden of Diseases Study 2015. Lancet 2017, 389, 1907–1918. [CrossRef] 7. Kinney, P.L. Climate Change, Air Quality, and Human Health. Am J. Prev. Med. 2008, 35, 459–467. [CrossRef] [PubMed] 8. Lelieveld, J.; Evans, J.S.; Fnais, M.; Giannadaki, D.; Pozzer, A. The contribution of outdoor air pollution sources to premature mortality on a global scale. Nature 2015, 525, 367. [CrossRef][PubMed] 9. Pannullo, F.; Lee, D.; Neal, L.; Dalvi, M.; Agnew, P.; O’Connor, F.M.; Mukhopadhyay, S.; Sahu, S.; Sarran, C. Quantifying the impact of current and future concentrations of air pollutants on respiratory disease risk in England. Environ. Health 2017, 16, 29. [CrossRef][PubMed] 10. Silva, R.A.; West, J.J.; Lamarque, J.-F.; Shindell, D.T.; Collins, W.J.; Faluvegi, G.; Folberth, G.A.; Horowitz, L.W.; Nagashima, T.; Naik, V.; et al. Future global mortality from changes in air pollution attributable to climate change. Nat. Clim. Chang. 2017, 7, 647–651. [CrossRef] 11. Kim, D.; Stockwell, W.R. An online coupled meteorological and air quality modeling study of the effect of complex terrain on the regional transport and transformation of air pollutants over the Western United States. Atmos. Environ. 2008, 42, 4006–4021. [CrossRef] 12. Grigoras, G.; Cuculeanu, V.; Ene, G.; Mocioaca, G.; Deneanu, A. Air pollution dispersion modeling in a polluted industrial area of complex terrain from Romania. Rom. Rep. Phys. 2012, 64, 173–186. 13. Lagzi, I.; Mészáros, R.; Gelybó, G.; Leel˝ossy, Á. Theory and limitations of Gaussian models. In Atmospheric Chemistry; Eötvös Loránd University-ELTE: Budapest, Hungary, 2013. 14. Lutman, E.R.; Jones, S.R.; Hill, R.A.; McDonald, P.; Lambers, B. Comparison between the predictions of a Gaussian plume model and a Lagrangian particle dispersion model for annual average calculations of long-range dispersion of radionuclides. J. Environ. Radioact. 2004, 75, 339–355. [CrossRef] 15. Seigneur, C.; Moran, M. Chapter 8. Chemical-Transport Models. Available online: https://www.narsto.org/ sites/narsto-dev.ornl.gov/files/Ch71.3MB.pdF (accessed on 6 December 2018). 16. NCAR. Models. Available online: https://www2.acom.ucar.edu/bai/models (accessed on 12 May 2018). 17. Ahmadov, R. WRF-Chem: Online vs Offline Atmospheric Chemistry Modeling. In ASP Colloquium; National Center for Atmospheric Research (NCAR): Boulder, CO, USA, 2016.

18. Feng, X.; Li, Q.; Zhu, Y.; Hou, J.; Jin, L.; Wang, J. Artiﬁcial neural networks forecasting of PM2.5 pollution using air mass trajectory based geographic model and wavelet transformation. Atmos. Environ. 2015, 107, 118–128. [CrossRef] 19. Jimenez, P.A.; Dudhia, J. On the Ability of the WRF Model to Reproduce the Surface Wind Direction over Complex Terrain. J. Appl. Meteorol. Climatol. 2013, 1610–1617. [CrossRef] 20. Baklanov, A. Application of CFD Methods for Modelling in Air Pollution Problems: Possibilities and Gaps. Environ. Monit. Assess. 2000, 65, 181–189. [CrossRef] 21. Chen, J.; Chen, H.; Wu, Z.; Hu, D.; Pan, J.Z. Forecasting smog-related health hazard based on social media and physical sensor. Inf. Syst. 2017, 64, 281–291. [CrossRef] 22. Grange, S.K.; Carslaw, D.C.; Lewis, A.C.; Boleti, E.; Hueglin, C. Random forest meteorological normalisation

models for Swiss PM10 trend analysis. Atmos. Chem. Phys. Discuss. 2018, 1–28. [CrossRef] 23. Liu, B.-C.; Binaykia, A.; Chang, P.-C.; Tiwari, M.; Tsao, C.-C. Urban air quality forecasting based on multidimensional collaborative Support Vector Regression (SVR): A case study of Beijing-Tianjin-Shijiazhuang. PLoS ONE 2017, 12, e0179763. [CrossRef] 24. Shimadera, H.; Kojima, T.; Kondo, A. Evaluation of Air Quality Model Performance for Simulating

Long-Range Transport and Local Pollution of PM2.5 in Japan. Adv. Meteorol. 2016, 2016.[CrossRef] 25. Ritter, M.; Müller, M.D.; Tsai, M.-Y.; Parlow, E. Air pollution modeling over very complex terrain: An evaluation of WRF-Chem over Switzerland for two 1-year periods. Atmos. Res. 2013, 132–133, 209–222. [CrossRef] 26. Lee, M.; Brauer, M.; Wong, P.; Tang, R.; Tsui, T.H.; Choi, C.; Cheng, W.; Lai, P.-C.; Tian, L.; Thach, T.-Q.; et al. Land use regression modelling of air pollution in high density high rise cities: A case study in Hong Kong. Sci. Total Environ. 2017, 592, 306–315. [CrossRef] Appl. Sci. 2018, 8, 2570 25 of 27

27. Nhung, N.T.T.; Amini, H.; Schindler, C.; Kutlar Joss, M.; Dien, T.M.; Probst-Hensch, N.; Perez, L.; Künzli, N. Short-term association between ambient air pollution and pneumonia in children: A systematic review and meta-analysis of time-series and case-crossover studies. Environ. Pollut. 2017, 230, 1000–1008. [CrossRef]

28. Zafra, C.; Ángel, Y.; Torres, E. ARIMA analysis of the effect of land surface coverage on PM10 concentrations in a high-altitude megacity. Atmos. Pollut. Res. 2017, 8, 660–668. [CrossRef] 29. Zhu, W.; Wang, J.; Zhang, W.; Sun, D. Short-term effects of air pollution on lower respiratory diseases and forecasting by the group method of data handling. Atmos. Environ. 2012, 51, 29–38. [CrossRef] 30. Zhang, Y.; Bocquet, M.; Mallet, V.; Seigneur, C.; Baklanov, A. Real-time air quality forecasting, part I: History, techniques, and current status. Atmos. Environ. 2012, 60, 632–655. [CrossRef] 31. Brokamp, C.; Jandarov, R.; Rao, M.B.; LeMasters, G.; Ryan, P. Exposure assessment models for elemental components of particulate matter in an urban environment: A comparison of regression and random forest approaches. Atmos. Environ. 2017, 151, 1–11. [CrossRef][PubMed] 32. Adam-Poupart, A.; Brand, A.; Fournier, M.; Jerrett, M.; Smargiassi, A. Spatiotemporal modeling of ozone levels in Quebec (Canada): A comparison of kriging, land-use regression (LUR), and combined Bayesian maximum entropy-LUR approaches. Environ. Health Perspect. 2014, 122, 970–976. [CrossRef][PubMed] 33. Hoek, G.; Beelen, R.; de Hoogh, K.; Vienneau, D.; Gulliver, J.; Fischer, P.; Briggs, D. A review of land-use regression models to assess spatial variation of outdoor air pollution. Atmos. Environ. 2008, 42, 7561–7578. [CrossRef] 34. Marshall, J.D.; Nethery, E.; Brauer, M. Within-urban variability in ambient air pollution: Comparison of estimation methods. Atmos. Environ. 2008, 42, 1359–1369. [CrossRef] 35. Zhan, Y.; Luo, Y.; Deng, X.; Grieneisen, M.L.; Zhang, M.; Di, B. Spatiotemporal prediction of daily ambient ozone levels across China using random forest for human exposure assessment. Environ. Pollut. 2018, 233, 464–473. [CrossRef] 36. Martínez-España, R.; Bueno-Crespo, A.; Timón, I.; Soto, J.; Muñoz, A.; Cecilia, J.M. Air-pollution prediction in smart cities through machine learning methods: A case of study in Murcia, Spain. J. Univ. Comput. Sci. 2018, 24, 261–276. 37. Bougoudis, I.; Demertzis, K.; Iliadis, L. HISYCOL a hybrid computational intelligence system for combined machine learning: The case of air pollution modeling in Athens. Neural Comput. Appl. 2016, 27, 1191–1206. [CrossRef] 38. Sayegh, A.S.; Munir, S.; Habeebullah, T.M. Comparing the Performance of Statistical Models for Predicting

PM10 Concentrations. Aerosol Air Qual. Res. 2014, 10, 653–665. [CrossRef] 39. Singh, K.P.; Gupta, S.; Rai, P. Identifying pollution sources and predicting urban air quality using ensemble learning methods. Atmos. Environ. 2013, 80, 426–437. [CrossRef] 40. Philibert, A.; Loyce, C.; Makowski, D. Prediction of N2O emission from local information with Random Forest. Environ. Pollut. 2013, 177, 156–163. [CrossRef][PubMed] 41. Nieto, P.J.G.; Antón, J.C.Á.; Vilán, J.A.V.; García-Gonzalo, E. Air quality modeling in the Oviedo urban area (NW Spain) by using multivariate adaptive regression splines. Environ. Sci. Pollut. Res. 2015, 22, 6642–6659. [CrossRef][PubMed]

42. Kleine Deters, J.; Zalakeviciute, R.; Gonzalez, M.; Rybarczyk, Y. Modeling PM2.5 Urban Pollution Using Machine Learning and Selected Meteorological Parameters. J. Electr. Comput. Eng. 2017, 2017.[CrossRef] 43. Carnevale, C.; Finzi, G.; Pederzoli, A.; Turrini, E.; Volta, M. Lazy Learning based surrogate models for air quality planning. Environ. Model. Softw. 2016, 83, 47–57. [CrossRef] 44. Suárez Sánchez, A.; García Nieto, P.J.; Riesgo Fernández, P.; del Coz Díaz, J.J.; Iglesias-Rodríguez, F.J. Application of an SVM-based regression model to the air quality study at local scale in the Avilés urban area (Spain). Math. Comput. Model. 2011, 54, 1453–1466. [CrossRef] 45. Vong, C.M.; Ip, W.F.; Wong, P.K.; Yang, J.Y. Short-term prediction of air pollution in Macau using support vector machines. J. Control Sci. Eng. 2012, 2012.[CrossRef] 46. Chen, S.; Kan, G.; Li, J.; Liang, K.; Hong, Y. Investigating China’s urban air quality using big data, information theory, and machine learning. Pol. J. Environ. Stud. 2018, 27, 565–578. [CrossRef] 47. Papaleonidas, A.; Iliadis, L. Neurocomputing techniques to dynamically forecast spatiotemporal air pollution data. Evol. Syst. 2013, 4, 221–233. [CrossRef] 48. Oprea, M.; Dragomir, E.G.; Popescu, M.; Mihalache, S.F. Particulate Matter Air Pollutants Forecasting using Inductive Learning Approach. Rev. Chim. 2016, 67, 2075–2081. Appl. Sci. 2018, 8, 2570 26 of 27

49. Liu, Y.; Cao, G.; Zhao, N.; Mulligan, K.; Ye, X. Improve ground-level PM2.5 concentration mapping using a random forests-based geostatistical approach. Environ. Pollut. 2018, 235, 272–282. [CrossRef][PubMed] 50. Just, A.; De Carli, M.; Shtein, A.; Dorman, M.; Lyapustin, A.; Kloog, I. Correcting Measurement Error

in Satellite Aerosol Optical Depth with Machine Learning for Modeling PM2.5 in the Northeastern USA. Remote Sens. 2018, 10, 803. [CrossRef] 51. Zhan, Y.; Luo, Y.; Deng, X.; Chen, H.; Grieneisen, M.L.; Shen, X.; Zhu, L.; Zhang, M. Spatiotemporal prediction

of continuous daily PM2.5 concentrations across China using a spatially explicit machine learning algorithm. Atmos. Environ. 2017, 155, 129–139. [CrossRef] 52. Xu, J.; Schussler, O.; Rodriguez, D.G.L.; Romahn, F.; Doicu, A. A Novel Ozone Proﬁle Shape Retrieval Using Full-Physics Inverse Learning Machine (FP-ILM). IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2017, 10, 5442–5457. [CrossRef]

53. De Hoogh, K.; Héritier, H.; Stafoggia, M.; Künzli, N.; Kloog, I. Modelling daily PM2.5 concentrations at high spatio-temporal resolution across Switzerland. Environ. Pollut. 2018, 233, 1147–1154. [CrossRef] 54. Hu, K.; Rahman, A.; Bhrugubanda, H.; Sivaraman, V. HazeEst: Machine Learning Based Metropolitan Air Pollution Estimation from Fixed and Mobile Sensors. IEEE Sens. J. 2017, 17, 3517–3525. [CrossRef] 55. Gacquer, D.; Delcroix, V.; Delmotte, F.; Piechowiak, S. Comparative study of supervised classiﬁcation algorithms for the detection of atmospheric pollution. Eng. Appl. Artif. Intell. 2011, 24, 1070–1083. [CrossRef] 56. Abu Awad, Y.; Koutrakis, P.; Coull, B.A.; Schwartz, J. A spatio-temporal prediction model based on support vector machine regression: Ambient Black Carbon in three New England States. Environ. Res. 2017, 159, 427–434. [CrossRef] 57. Araki, S.; Shima, M.; Yamamoto, K. Spatiotemporal land use random forest model for estimating metropolitan

NO2 exposure in Japan. Sci. Total Environ. 2018, 634, 1269–1277. [CrossRef] 58. Beckerman, B.S.; Jerrett, M.; Martin, R.V.; van Donkelaar, A.; Ross, Z.; Burnett, R.T. Application of the deletion/substitution/addition algorithm to selecting land use regression models for interpolating air pollution measurements in California. Atmos. Environ. 2013, 77, 172–177. [CrossRef]

59. Yang, W.; Deng, M.; Xu, F.; Wang, H. Prediction of hourly PM2.5 using a space-time support vector regression model. Atmos. Environ. 2018, 181, 12–19. [CrossRef] 60. Li, L.; Lurmann, F.; Habre, R.; Urman, R.; Rappaport, E.; Ritz, B.; Chen, J.C.; Gilliland, F.D.; Wu, J. Constrained Mixed-Effect Models with Ensemble Learning for Prediction of Nitrogen Oxides Concentrations at High Spatiotemporal Resolution. Environ. Sci. Technol. 2017, 51, 9920–9929. [CrossRef][PubMed] 61. Zhang, J.; Ding, W. Prediction of air pollutants concentration based on an extreme learning machine: The case of Hong Kong. Int. J. Environ. Res. Public Health 2017, 14, 1–19. [CrossRef][PubMed] 62. Peng, H.; Lima, A.R.; Teakles, A.; Jin, J.; Cannon, A.J.; Hsieh, W.W. Evaluating hourly air quality forecasting in Canada with nonlinear updatable machine learning methods. Air Qual. Atmos. Health 2017, 10, 195–211. [CrossRef] 63. Yeganeh, B.; Motlagh, M.S.P.; Rashidi, Y.; Kamalan, H. Prediction of CO concentrations based on a hybrid Partial Least Square and Support Vector Machine model. Atmos. Environ. 2012, 55, 357–365. [CrossRef] 64. Tamas, W.; Notton, G.; Paoli, C.; Nivet, M.L.; Voyant, C. Hybridization of air quality forecasting models using machine learning and clustering: An original approach to detect pollutant peaks. Aerosol Air Qual. Res. 2016, 16, 405–416. [CrossRef]

65. Ni, X.Y.; Huang, H.; Du, W.P. Relevance analysis and short-term prediction of PM2.5 concentrations in Beijing based on multi-source data. Atmos. Environ. 2017, 150, 146–161. [CrossRef] 66. Li, C.; Zhu, Z. Research and application of a novel hybrid air quality early-warning system: A case study in China. Sci. Total Environ. 2018, 626, 1421–1438. [CrossRef] 67. Eldakhly, N.M.; Aboul-Ela, M.; Abdalla, A. A Novel Approach of Weighted Support Vector Machine with Applied Chance Theory for Forecasting Air Pollution Phenomenon in Egypt. Int. J. Comput. Intell. Appl. 2018, 17, 1–29. [CrossRef] 68. Wang, D.; Wei, S.; Luo, H.; Yue, C.; Grunder, O. A novel hybrid model for air quality index forecasting based on two-phase decomposition technique and modiﬁed extreme learning machine. Sci. Total Environ. 2017, 580, 719–733. [CrossRef][PubMed] 69. Zhao, X.; Zhang, R.; Wu, J.-L.; Chang, P.-C. A deep recurrent neural network for air quality classiﬁcation. J. Inf. Hiding Multimed. Signal Process. 2018, 9, 346–354. Appl. Sci. 2018, 8, 2570 27 of 27

70. Huang, C.J.; Kuo, P.H. A deep cnn-lstm model for particulate matter (PM2.5) forecasting in smart cities. Sensors 2018, 18.[CrossRef] 71. Sadiq, A.; El Fazziki, A.; Ouarzazi, J.; Sadgal, M. Towards an agent based trafﬁc regulation and recommendation system for the on-road air quality control. Springerplus 2016, 5.[CrossRef] 72. Tzima, F.A.; Mitkas, P.A.; Voukantsis, D.; Karatzas, K. Sparse episode identiﬁcation in environmental datasets: The case of air quality assessment. Expert Syst. Appl. 2011, 38, 5019–5027. [CrossRef] 73. Debry, E.; Mallet, V. Ensemble forecasting with machine learning algorithms for ozone, nitrogendioxide and

PM10 on the Prev’Air platform. Atmos. Environ. 2014, 91, 71–74. [CrossRef] 74. Shaban, K.B.; Kadri, A.; Rezk, E. Urban Air Pollution Monitoring System with Forecasting Models. IEEE Sens. J. 2016, 16, 2598–2606. [CrossRef]

75. United States Environmental Protection Agency. Particulate Matter (PM2.5) Trends. Available online: https://www.epa.gov/air-trends/particulate-matter-pm25-trends (accessed on 5 December 2018). 76. European Environment Agency. Air Quality in Europe—2018 Report; European Environment Agency: Copenhagen, Denmark, 2018. 77. European Environment Agency. Air Quality in Europe—2017 Report; European Environment Agency: Copenhagen, Denmark, 2017. 78. Zalakeviciute, R.; Rybarczyk, Y.; López-Villada, J.; Diaz Suarez, M.V. Quantifying decade-long effects of fuel

and traffic regulations on urban ambient PM2.5 pollution in a mid-size South American city. Atmos. Pollut. Res. 2017.[CrossRef] 79. Miller, M.R.; Raftis, J.B.; Langrish, J.P.; McLean, S.G.; Samutrtai, P.; Connell, S.P.; Wilson, S.; Vesey, A.T.; Fokkens, P.H.B.; Boere, A.J.F.; et al. Inhaled Nanoparticles Accumulate at Sites of Vascular Disease. ACS Nano 2017, 11, 4542–4552. [CrossRef][PubMed] 80. Al–Dabbous, A.N.; Kumar, P.; Khan, A.R. Prediction of airborne nanoparticles at roadside location using a feed–forward artificial neural network. Atmos. Pollut. Res. 2017, 8, 446–454. [CrossRef] 81. Pandey, G.; Zhang, B.; Jian, L. Predicting submicron air pollution indicators: A machine learning approach. Environ. Sci. Process. Impacts 2013, 15, 996–1005. [CrossRef][PubMed] 82. Prank, M.; Sofiev, M.; Tsyro, S.; Hendriks, C.; Semeena, V.; Francis, X.V.; Butler, T.; Van Der Gon, H.D.; Friedrich, R.; Hendricks, J.; et al. Evaluation of the performance of four chemical transport models in predicting the aerosol chemical composition in Europe in 2005. Atmos. Chem. Phys. 2016, 16, 6041–6070. [CrossRef]

© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).